Thriller, sex, drama, surveillance, or what it takes to be an engineer in the 21st century

I have the most typical life. To fully appreciate how typical my life is I will start from the very beginning.

I never appreciated learning, because I went to school too early, at the age of five. My mother raising us alone put me and my older brother in the same grade to match his age — so that he could keep an eye on me while she was working. I was terrible at my classes. I could not understand why I needed to do what my teacher told me to do and not do what I wanted to — read the books that I wanted, solve the problems that I wanted. My internal voice and my ego drove me away from the teachers’ agenda from the very childhood. I cannot explain it but all my life I thought I knew better what I needed. Maybe because the teachers were forced to focus less on sparking curiosity and more on keeping up with a rigid, mandated curriculum.

To keep me busy and entertained I did a lot of extra curricular activity like chess, photography, aircraft modeling, karting, and later sports — basketball and football. My mother wanted us to have a full and diverse childhood despite of our father’s early death by the heart attack and us witnesing the war in Abkhazia in the 90s.

So at the age of 13, I decided to concentrate on sports rather than on education and started playing football professionally (children-professionally). Twice we were vice-champions of Moscow Second Youth Football League. Scoring regularly, even in important matches like a cup game with Locomotive, I was under impression that I will become a professional football player.

Well, it did not happen. At the age of 15 my hormones started to kick in, and, in high school, I quickly realized that I no longer could do sports well, neither I could concentrate on my studies. To make it even worse I joined a rock band as a solo guitarist. A typical teenager constantly thinking about sex.

Surprisingly, love (and a book about life of soviet theoretical physicist Lev Landau and his numerous mistresses, which I read 3 times) saved me. I met a very beautiful and smart girl highly valuing education. As a typical boy, I just wanted to impress her and the only way to impress her was, well, by becoming smart. Long story short, I successfully entered Moscow Aviation Institute just in time, because of luck, a love with the right person, and really got interested in science and math since then.

To be fully accurate, love was not the only reason why I decided to concentrate on engineering at college. I think it was predetermined by two reasons: first, my mother always spoke about the value of higher education and, second, my father was a soviet engineer working in a secret lab, or at least this is what my mother told us. In other words, you could say me becoming an engineer was written in the book of fate.

I got so excited that I even completed my PhD in mathematical modelling in 2012 at the Department of Physics (different department from my bachelor and masters). And that is where our story begins.

Fresh out of college but with the doctorate degree, I decided not to pursue academia but started my career in industry instead. I joined the largest offshore construction company called MRTS, or Mezhregiontruboprovodstroy, in 2012, right at the dawn of offshore golden age in Russia.

To show me the ropes, managements decided to send me to the pipelay vessel in the Sea of Okhotsk to get acquainted with the profession. I was so excited, that I could not sleep well. I got even more excited when I got on the barge and realized the magnitude of the construction site. I remember the serious atmosphere of being on a barge at sea. There was a big slogan written on deckhouse — Safety Comes First. I also remember that colleagues asked me to smuggle vodka in. But cannot remember if I did so :)

The biggest outcome of that trip was that I did calculation for pipeline installation and thus helped to complete a real offshore procedure. Not sure if I did that calculation 100% right but the procedure went successfully.

Upon returning to Moscow office, I already had a name in the company, and the company tasked me with the next challenge. Namely, to create an in-house methodology for offshore pipelines installation analysis. Again, I got so excited that I verified that methodology with Det Norske Veritas around 2013. It shut everybody down. I was promoted to the Head of Engineering Calculation Division and started building the division from scratch by negotiating purchases of OrcaFlex, Ansys, SolidWorks, and other software. At the same time, my desire to learn prevailed, and I successfully lobbied for the team to take course on installation calculations from a British company. At that time we were doing a lot of international activity like visiting partners in Denmark and being invited to Norwegian embassy (which I had ignored). I even got interviewed about the future of arctic in 2013 or 2014. And then, we all know what happened in Russia in 2014 when all our international activity started to come to the end.

Remembering that time, I participated, as a key engineer, in all main projects in Russian offshore from 2012 to 2016 like Kirinskoye, Varandey Terminal, Novoportovskoye, Nord stream — 2, Turkish Stream, South-Kirinskoye, etc. It was a good time that I only now begin to realize.

But it ended ugly for me. With all the attention and significance of the projects I worked on, I lost sleep, grew mildly paranoid, and had what I can only call a nervous breakdown. In 2016, I quit my job. My fiancée left me. And I slipped into what I would describe as a low-grade depression.

The life events were so intense that I could not make sense out of it anymore. And what I decide? Well, I decide to become a data scientist. The best decision I have ever made because of my brother who introduced me to the profession.

In just one year, I mastered the profession on my own and found my first job as a data scientist. But even though I left engineering, engineering did not leave me. Around 2017, I started thinking on combining my old profession of mechanical engineer and my new profession of AI scientist for my personal projects outside corporate work. I got lucky again and the right book found me. It was Reinforcement Learning: An Introduction by Sutton & Barto. I immediately realized that I needed to apply RL to mechanical engineering. Believe it or not but I rediscovered topology optimization on my own. I simply thought that combining finite element methods as an environment and reinforcement learning as an agent was a good idea. I set up my GitHub repo and made my first commit to ‘Reinforcement_Learning_and_FEA’ (later Gigala) in October 2018. By March-June 2019, the repo had a grid-based design agent, well before the first peer-reviewed papers, and was noticeably popular with 10–15 stars. That is how I became, as far as I can tell, one of the first people to work on topology optimization by reinforcement learning in public [1–4].

If you think this is the end of my story, well, you are wrong, it is just a beginning.

In late 2018, I went to China to work in Artificial Intelligence Town in the city of Hangzhou. I also shared some insight on my personal project with my colleague Ollie who encouraged me to continue my efforts.

It is hard to realize importance of your work if you are the only person in the world working on the problem. I could not say that there was no winter in my research in TO by RL. For example, in 2019, I digressed and decided to develop the app that helped in creating engaging posts to share on social media. You just needed to input a short sentence describing your thoughts and provide a path to a photo album. The app selected the photo that reflected your mood the best and created an engaging caption to share both on a social media platform. You know, the Facebook style startuping. I even hosted it online via Google Cloud for a while. I had a slogan — “just gigala it”, meaning just create a post with my app. Back then I used Gigala word to mean something different than TO by RL. But all winters come to the end. Occasionally continuing working on topology optimization by reinforcement learning, I got a message from my ex-boss asking if I wanted to return to offshore construction around 2020. Only many years later I realized that that request might have been linked to Nord Stream — 2 project.

So what did I do, I decided what a good idea was to make some extra money by making calculations for pipeline installation. I rush to buy a licence for OrcaFlex. Because of my passport, I am denied to do so. What do I do, I start developing my own software for offshore pipeline installation analysis called Ocean Intella in around 2021. Five years later the work on this software paid off.

In 2022, the winter of topology optimization by reinforcement learning for me was over. I read Nathan Brown et al, and Hayashi & Ohsaki papers. Suddenly, I realized that the interest in community took off. Not to get left behind, I took my will in my hands and wrote a couple of peer reviewed papers.

It is a cosmic coincidence that Hayashi & Ohsaki used the same number of elements as me, namely 72, only 6–12 months later than me when my preprints already were a part of a public discourse on my github (see e.g. [5]).

Importantly, I make no claim that Hayashi and Ohsaki had prior awareness of my work. Only about my historical priority.

Since then an interest in my work gradually rose. I cannot remember all, but a Lockheed Martin engineer, American and Canadian students, a researcher on fusion, a researcher in chip design, ESA, angel investors, and even a White House AI Strategist demonstrated an interest in my work on Gigala.

Oh, yes, in 2026, one of my papers in topology optimization by reinforcement learning got cited in Nature Portfolio, [6].

That is my story.

At the end of my narrative, you need some words of wisdom. Something to pass to the next generations. Well, here they are:

I hate science. I hate it because I do it everyday. When you do science everyday you learn how to hate it — since most of the time, it does not work. I mean 99.99% of the time you do not get the result. I love science. I love science because it gave me meaning when I was at the lowest in my life. But I also love science because it is one of the rare things that we know are important.

[1]https://github.com/gigatskhondia/gigala/commit/889388508a03129365da55db8e8ca1048ac29ab8

[2]https://github.com/gigatskhondia/gigala/tree/eda024aa734c1014b24a65303832ed92687b0ed1

[3]https://github.com/gigatskhondia/gigala/tree/88fcfb91027433dfcffe165e5e3ab6dc2b2e1926

[4] https://github.com/gigatskhondia/gigala/commit/9bb730a002baf92765ebb6d8d78b14621bfd481b

[5]https://github.com/gigatskhondia/gigala/blob/b97b4f0af52998d992a77f8680f1f9a68b958187/manuals/manual_bionic_partition_AlphaZero.pdf

[6] https://www.nature.com/articles/s41598-026-35863-1

To keep up with the project please visit https://gigala.io/

In my recent article on Medium, [1], I introduced a concept of “agentic metals”. This concept stands that there is some “agency” in metal alloys that allows them to control dynamics of dislocations. But recently I thought — what if we go a bit further? What if we introduce “Agentic Physics”? Here by agents I mean reinforcement learning (RL). What if every physical process is a Markov Decision Process (MDP) that performs actions of applying a physical force to the environment? Now physics would equal a neural network (NN) which we get after training the RL agent. The reward signal could be a physics simulator (in a traditional sense) or physical experiment. We could make this NN as custom as we want to as custom of a physical process as we want. All of this would open a huge possibility of discovering new physics laws by transfer learning, where a new physical process is generalized by a “close sample” from “the training set”. In the future there will not be a discovery of a new physics law there will be training and generalizing “physics neural networks”.

Let’s consider the case where Newton physics brakes and Einstein physics stands in — The Perihelion of Mercury — to exemplify the discovering process of a new physics by transfer learning.

Let’s ask claude.ai to produce us a nice description on The Perihelion of Mercury phenomenon.

Mercury’s orbit around the Sun slowly rotates over time — a phenomenon called orbital precession. The point of Mercury’s closest approach to the Sun (its perihelion) shifts forward with each orbit, tracing a slow arc around the Sun.

What Newton Predicts:

Newtonian gravity accounts for most of this precession through the gravitational tugs of other planets. Crunching the numbers with Newton’s laws, astronomers in the 19th century predicted a total precession of about 5,557 arcseconds per century.

What astronomers actually measured: 5,600 arcseconds per century

That leaves an unexplained gap of 43 arcseconds per century — tiny, but stubbornly real. For decades, no one could explain it. Some even proposed a hidden planet (“Vulcan”) orbiting close to the Sun to account for the discrepancy.

What Einstein Predicts:

General Relativity adds a correction: spacetime near a massive object like the Sun is curved, and that curvature causes an extra orbital shift Newton simply does not account for. Einstein’s equations predict exactly 43 additional arcseconds per century.

The match is essentially perfect.

The intuition: In Newton’s universe, a planet orbiting a perfectly spherical, isolated star would trace a fixed ellipse forever. Einstein says no — the warping of spacetime near the Sun means the ellipse itself slowly rotates. The more massive and close the star, the more dramatic the effect. Mercury, being the closest planet to the Sun, feels this most acutely.

This was actually one of the first major confirmations of General Relativity — and Einstein reportedly said he was so excited when his calculation matched observation that he had heart palpitations.

Now how would we discover this General Relativity by our Agentic Physics.

We would had a classical simulator with Newton physics applied to a total precession’s calculation of Mercury. At once we got the neural network we would change the applied force a bit so that it coincided with experiment and hence a new (below) formula could be “discovered” (not sure that this is the right formula for this example, but you get the picture):

Hence we would get at inference a new physics low by transfer learning — General Relativity Neural Network.

As to formalization, the action could be a small increment of the force applied to body/ environment, and the reward signal could be if force-action coincides with the force from experiment than — reward.

In other words:

State (s) — current physical configuration of the system

Action (a) — small incremental force applied to body/environment

Reward (r) — coincidence between predicted force and experimentally observed force

Policy (π) — the learned physics law itself

More formally, for example:

s(t+1) = f(s(t), a(t))
a(t) = δF (infinitesimal force increment)
r(t) = -|F_predicted - F_experimental|
π* = argmax E[Σ r(t)]

Transfer learning gets triggered if we slightly change the environment. For example, we could change the position of a body in the system or a number of bodies. Maybe, if we alter the environment it could lead to more general principals.

Now physics would become as an inverse problem solved by RL:

Instead of:

• Observe → Hypothesize → Derive equations → Test

We do:

• Observe experimentally → Define reward as match to observation → Let agent discover the force law → Extract transferable neural network

Let’s ask claude.ai to formalize it further.

The perturbation hierarchy:

Level 1 — Change body position

• Mercury at different orbital distances
• Agent must generalize force laws across spatial configurations
• Discovers: inverse square relationships, field geometry

Level 2 — Change number of bodies

• Two body → three body → n body problem
• Agent must generalize across system complexity
• Discovers: superposition principles, emergent interactions

Level 3 — Change environmental constants

• Alter mass, charge, velocity
• Agent generalizes across parameter space
• Discovers: dimensional relationships, conservation laws

Level 4 — Change physical regime entirely

• Classical → relativistic → quantum
• Transfer learning bridges regimes
• Discovers: unifying principles between theories

I think the state could be minimum parameters needed to describe the physical system because at first we might try to find new relations between known parameters rather than find new parameters themselves.

Let’s ask what claude.ai thinks on this.

Instead of trying to discover everything from scratch — which is computationally intractable:

Start with what we know. Find new relationships between known parameters first.

Let’s use physics’ own parsimony principle — Occam’s razor — as the state space constraint.

What “minimum parameters” means in practice

For Mercury perihelion:

• Mass of Sun
• Mass of Mercury
• Orbital radius
• Velocity
• Time

That’s it. Five parameters. Completely tractable.

For quantum systems:

• Energy levels
• Spin
• Charge
• Planck’s constant

Still finite. Still learnable.

This creates a beautiful two phase discovery process:

Phase 1 — Relational Discovery

• State = known minimum parameters
• Agent discovers new relationships between them
• Example: discovers relativistic correction to Newtonian gravity
• This is where Agentic Physics starts

Phase 2 — Parameter Discovery

• Anomalies in Phase 1 that can’t be explained by known parameters
• Signal that a new parameter is needed
• Agent flags the gap
• Human physicists investigate what new parameter might explain it
• Example: discovering dark matter as a parameter

I think, the number of minimum parameters could be found in the simulator iteratively, and the number that best matches the physical experiment would be adopted.

Let’s ask what claude.ai thinks on this.

He suggests to formalize it like this:

For n = 1 to N parameters:
    Define state s = {p1, p2, ... pn}
    Train RL agent with:
        - Action: incremental force δF
        - Reward: match to physical experiment
    Evaluate fit to experimental data
    
Adopt minimum n with best experimental match

Apply environmental perturbations:
    - Change body positions
    - Change number of bodies
    - Change regime

Trigger transfer learning across perturbations

Flag anomalies where no n gives sufficient match
→ Signal: new parameter needed

Sorry for such a crude example, but I hope you get the picture now of how Agentic Physics should work. Let’s work on the details of my theory together.

To keep up with the project please visit https://gigala.io/

[1] https://gigatskhondia.medium.com/ideation-of-relation-of-plastic-strain-rate-and-dislocations-and-a-concept-of-agentic-metal-b6ab27ce62c0

This article proposes a relation of plastic strain rate and dislocations in metals and stands how to derive it experimentally. New concept of “agentic metal” is presented.

Consider these ideas as flight of fancy that came to me one lonely evening.

Concept

Let’s assume the relation between plastic strain rate and dislocations:

Let’s derive the material constants from two experiments:

Experiment 1:

-polycrystalline alloy;

- we stretch the specimen with acceleration so that at the moment of material failure the deformation rate was maximum;

-at the moment of material failure (a split second before), we assume that Dρ=1, Ḋρ=1, hence:

Experiment 2:

- monocrystal (Dρ=D1), where D1 is a normalized dislocation density in monocrystal; typically in monocrystal a non-normalized dislocation density is 10⁶

- we stretch with acceleration, so that at the moment of material failure the deformation rate is maximum (but the maximum deformation rate may differ from experiment 1, hence the stroke in the formula):

- We assume in the end the dislocation density was close to D1 while a normalized dislocations rate was close to 1, and a failure immediately occurred.

Thus, from experiment 1 and 2 we find the constants C1 and C2.

Now starts the most interesting part.

A concept of “agentic metal”

Let’s assume that metal alloy has some “agency”, meaning that metal controls dislocation rate based on external loads. Let’s consider the same formula (1):

But now:

Dρ - random field of dislocations in the volume (area) of the specimen;

Ḋρ - actions performed by the “metal agent” (add a little, subtract a little) so that plastic strain rate corresponds to a macro-experiment or macro-simulation (if so — reward the RL agent).

Let’s assume that dislocations flow toward the maximum gradient of dislocations density. That is, with each action, the agent selects a point on the specimen with the maximum dislocation density and removes a portion of the dislocations from it toward the maximum gradient, by the amount of the agent’s “action.”

Agent is in the sense of reinforcement learning.

Features for the RL agent are [x, y, Dρ, P, strain rate]

x, y — coordinates

P — external loads

strain rate - rate of stretching of the specimen

Actions: +0.1; +0.2; +0.3; ……; +1.0 to dislocation density.

The dislocations flow toward the maximum gradient of dislocations density. That is, with each action, the agent selects a point on the specimen with the maximum dislocation density and removes a portion of the dislocations from it toward the maximum gradient, by the amount of the agent’s “action.”

By training the agent via computer simulations, we will be able to predict the “next experiment’s stress-strain curve” (which is slightly different from training simulations) due to so called transfer learning.

Conclusion

I have just presented some ideas on the relation between dislocations and plastic strain rate. I have also introduced the concept of “agentic metal” — a paradigm shift occurs if we think that metals have some “agency” to control the rate of change of dislocation density based on external influence.

All above ideas should be considered as preliminary and requiring refinement. Hope you had intellectual fun reading them.

To keep up with the project please visit https://gigala.io/

The era of RL assisted metal forming

Co-authored with Claude

My PhD thesis was dedicated to an investigation of innovative techniques that lead to the improved plasticity and maintained strength of metal alloys, i.e. so called Electroplasticity Effect.

Electroplasticity Effect occurs during the process of plastic deformation when a high electric current of certain duration is applied to metal specimens. It leads to an improved plasticity and preserved strength — making it easier to plastically deform. It is a very useful technology in aerospace industry with the application e.g. to aluminium and titanium alloys [1–2].

Three competing mechanisms explain this:

• Joule heating — the current heats the material, softening it thermally
• Electron-wind force — moving electrons exert a drag force on dislocations, helping them move
• De-pinning of dislocations from paramagnetic obstacles — the electromagnetic field liberates dislocations that are stuck

My recent year passion, however, is topology optimization by reinforcement learning [3–5]. One can think of topology optimization by reinforcement learning as of an intelligent design. You can design machinery’s elements and components or you can design bridges and buildings.

An obvious extension of my research would be a combination of these domains — i.e. applying RL to the effect of electroplasticity.

For example, RL + topology optimization + electroplasticity-informed manufacturing.

Here how it would work. The rolling mill is now a robot (Fig 1); it automatically adapts to different alloys, pulse impacts, and rolling conditions, eliminating the need for manual adjustments. This can be very useful in manufacturing. How this is done: we have AI (specifically, RL), and we train it in a simulator, for instance, optimizing for maximum material plasticity (while preserving the strength). No data is needed. The simulator uses the finite element method. We create a controller (RL) on the computer and then transfer it to the real world (physical rolling mill).

Electroplastic rolling is currently a highly manual, parameter-heavy process. An operator or engineer has to tune pulse frequency, current density, pulse duration, rolling speed, reduction ratio, temperature, and inter-pass timing — all simultaneously, all dependent on the specific alloy being processed. This is exactly the kind of high-dimensional, coupled, nonlinear control problem that RL was built for. Hence, RL assistance should allow to automate this process.

How the System Would Actually Work

The Simulator (Training Environment)

The FEM simulator is the heart of it. You would build a coupled electro-thermo-mechanical FEM model . The RL agent observes the current state — roll force feedback, surface temperature, electrical measurements, pass number, alloy identity — and outputs actions: pulse parameters, rolling speed, reduction per pass, inter-pass intervals. The reward function is the critical design choice. Something like:

Maximize elongation-to-failure (plasticity proxy) subject to: yield strength ≥ target, surface integrity constraints, energy consumption penalty.

Topology Optimization Layer

Here is where it gets particularly interesting and where my TO with RL work enters. The rolling mill is not just a controller problem — the geometry of the rolls themselves, and potentially the geometry and distribution of electrode contacts delivering the current pulses, can be topology-optimized. You could use TO to design:

• Roll profile geometries that distribute stress and current more uniformly across complex cross-sections
• Electrode contact geometries that produce the most favorable current density distribution in the workpiece
• Die or guide geometries optimized jointly for mechanical and electrical boundary conditions

Why This Is Industrially Compelling

The addressable problem is real and large. Aerospace manufacturers working with titanium alloys, nickel superalloys, and advanced aluminum alloys spend enormous effort on multi-pass rolling schedules. Getting those schedules wrong means scrapped material, which in titanium means very expensive scrap. An RL controller that automatically adapts to alloy-by-alloy variation and self-optimizes the pulse parameters could:

• Reduce forming force requirements by a large margin
• Enable forming of alloys currently considered too brittle for conventional rolling
• Eliminate the expert-knowledge bottleneck in schedule design
• Generalize across alloys without re-engineering the process from scratch

Alloy-agnostic generalization — is probably the most commercially valuable claim, because it is what eliminates the manual adjustment problem.

The era of RL assisted metal forming has just begun. Jump on!

To keep up with the project please visit https://gigala.io/

[1] Simulating the effect of a high density electric current pulse on the stress field during plastic deformation. GA Tskhondiya, NN Beklemishev — International journal of material forming, 2012 — Springer

[2] Modeling the effect of multiple pulse treatment on deformation processing. GA Tskhondiya — Journal of Physics: Conference Series, 2009 — iopscience.iop.org

[3] https://jngr5.com/index.php/journal-of-next-generation-resea/article/view/95

[4]https://www.sciencepublishinggroup.com/article/10.11648/j.ajasr.20261201.11

[5] https://github.com/gigatskhondia/gigala

Topology optimization by reinforcement learning (TO by RL) represents a paradigm shift in structural design: an intelligent agent — powered by a neural network — learns to sequentially place or remove material within a design domain, iteratively refining the layout to achieve optimal performance under given loads, boundary conditions, and constraints. Unlike classical mathematical methods that rely on gradients and assumptions of continuity, RL treats the problem as a game of strategic decisions, enabling the discovery of highly complex, non-intuitive topologies that traditional approaches often miss.

One can think of topology optimization by reinforcement learning as of an intelligent design. You can design machinery’s elements and components or you can design bridges and buildings. But now this approach transforms to something new.

You see, the thing is that TO by RL has a neural network as its component. The neural network makes decision how to assemble a structure step by step in a sequence. But the idea that recently came to my mind was why not to transfer that neural network into physical domain?! Why not to put it, for example, in the brain of a robot?! And now suddenly instead of a design algorithm you have an assembly algorithm. You get what I call agent-technologists and agent-builders who use the ‘brain’ or the neural network obtained by topology optimization with RL to build in physical world.

To elucidate, think of a natural disaster - a hurricane or an earthquake. It damages buildings, roads, factories, shops, etc. However, now, you just give an agent-builder the task to repair the damage. You photo damaged building, put the photo in your computer. It is an initial topology of your structure. You set proper boundary conditions for the algorithm and do TO by RL to suggest repairs so that your structure is strong and light again. After you got the sequence of repair steps, you put the very neural network that made decisions to assemble the topology, in the robot’s brain, and just like that, the robot is ready to repair the building in physical world.

Another example would be space exploration. Think of lunar bases, for instance. It is dangerous to send people to assemble the lunar base, but you could send robots who would carry both topology optimization algorithm and assembly algorithms inside their brains. The robot would arrive to the moon, study terrain, choose a place for the base, photo or video record the surroundings, design the structure with TO by RL, and then assemble that structure on the terrain by the sequence of assembly actions provided from reinforcement learning design. The robot (or swarm of robots) would use lunar regolith and other available materials to support the construction of the structure. Hence, agent-builder paradigm could meaningfully contribute to colonization of other worlds (Fig. 1).

Having worked on TO by RL since 2018 and published several papers, [1–2], I continue pushing the boundary of what is possible with this approach. For example, my recent anchor node strategy (Fig. 2) and full 3D TO with RL (Fig. 3) informs what I am doing currently.

Thus, TO by RL excels at multi-objective optimization, delivering structures that simultaneously maximize strength while dramatically minimizing weight — the holy grail for aerospace, automotive, and space applications. It scales naturally to 3D problems via e.g. pseudo 3D approach, adapts to irregular constraints, and, crucially, produces policies that can be transferred directly into the physical world. The same neural network that designs the structure can guide a swarm of robots to actually build it, turning digital optimization into real-world autonomous construction.

This matters because raw material performance alone is not enough. Cutting-edge nanomaterials — no matter how strong or lightweight at the atomic scale — cannot deliver their promise without an intelligent macro- or meso-scale architecture. Even advanced composites fall short if their internal structure is suboptimal; they still require a carefully engineered topology to channel forces, avoid stress concentrations, and achieve the full strength-to-weight potential.

In sum, one can say that topology optimization with reinforcement learning has achieved preindustrial grade where it is able to design simple components like pins, hinges or small elements. And all of this is possible on your personal computer within 1hr wall-clock run time. If in doubt, check my open-source software Gigala, [3], and start building today.

Updated May 5, 2026:

I was able to achieve 20x20 grid with pure DRL via using progressive RL refinement where I initially trained the model in a usual way and got the topology that looked like Fig. 4, and then I trained the model again only for adjacent to that topology nodes and got something that looked like Fig. 5. Importantly, 20x20 grid is 2⁴⁰⁰ states which, to the best of my knowledge, is a new SOTA. To sum, I believe that with better refinement strategies more enhanced topologies should be achieved.

Updated May 19, 2026:

For 20x20 grid, I used “post-processing” where I initially trained the model in a usual way and got the topology that looked like Fig. 6, and then I applied box blur, topology filtering, and thresholding and got something that looked like Fig. 7.

Updated Sep 25, 2026:

With a better hyperparameter tuning and longer training I was able to get enhanced topologies for 20x20 grid (Fig.8).

Results on Figure 8 hint on potential scalability of my algorithm to bigger resolutions with more training and better hyperparameters’ tuning. Adopting my algorithm to run on GPU might give a potential boost in this direction.

To keep up with the project please visit https://gigala.io/

[1] https://jngr5.com/index.php/journal-of-next-generation-resea/article/view/95

[2]https://www.sciencepublishinggroup.com/article/10.11648/j.ajasr.20261201.11

[3] https://github.com/gigatskhondia/gigala

A prerequisite for ‘anchor node’ strategy?

I do topology optimisation by reinforcement learning research; for fun; in my free time. You can read more about my endeavours in my blog on Medium or in [1–3]. The journey is bumpy and with a lot of challenges, but I like it this way.

As I said before, an intriguing speculative idea struck me while watching the NVIDIA Keynote at COMPUTEX 2025, where Jensen Huang mentioned that “we are rendering only one out of ten pixels and guessing the remaining nine” using a technology called DLSS. This concept of selectively computing and inferring the rest inspired me to think on applying a similar principle to topology optimisation. Instead of calculating the entire topology, what if we compute only a subset of key “anchor” nodes and use a neural network to infer the remaining structure?! The combined layout — consisting of anchor and predicted nodes — could then be evaluated using finite element analysis (FEA) to assess structural compliance, providing feedback to guide the optimization process. This approach would have the potential to significantly accelerate learning and improve computational efficiency.

In this article, I wanted to discuss some progress that I am making on the road to my “anchor node” strategy.

I have called my new approach “GA pre-training RL parallel block-refinement” strategy (GA-pt-RL-pbr). Please check my GitHub for more implementation details (https://github.com/gigatskhondia/gigala/). But it goes like this: you divide the grid into blocks. For example, 28 by 28 grid you can divide into 16 equal blocks of 7 by 7 each. Then, first you do GA pre-training on the full 28x28 grid, after that you do RL refinement for each of 7x7 blocks in parallel. You just freeze the rest (taking topology from GA ) and refine one 7x7 block at a time (but in parallel). To do it in parallel you freeze and refine different parts. And the most interesting thing is that if you look at the pictures below it might be just the stuff I was looking for for my so called ‘anchor node’ strategy, [2].

Here is a topology after GA pre-training (Fig. 1).

And here what I get after the RL refinement stage (Fig. 2).

And finally, if we connect the dots, we get Fig. 3.

As you can see, topology on Fig.3 resembles the topology on Fig. 1.

More broadly, the ‘GA pre-training RL parallel block-refinement’ strategy can potentially be used to find topologies for bigger design spaces or to train generative models where, in terms of machine learning, you have data (Fig.2) and target (Fig.1 or Fig.3) for different boundary conditions. In fact, the nodes on (Fig. 3) are positioned at places of minimum compliance, hence the structure might be seen as a light weight product of generative design.

To sum up, I just wanted to say that my research on topology optimisation by reinforcement learning is not just optimal shapes. It is forming brains for agent-builders and agent-technologists. After an agent learns an optimal design, it has the brain (neural network) to implement that design in physical world i.e. to build a house, a bridge, or a road in a sequential manner. Applying my methods to robotics, where my algorithms provide the brain of the agent-builder, opens whole new possibilities for future constructions.

Codebase availability: https://github.com/gigatskhondia/gigala

To keep up with the project please visit https://gigala.io/

[1] https://www.researchgate.net/publication/389785241_Reinforcement_Learning_Guided_Engineering_Design_from_Topology_Optimization_to_Advanced_Modelling

[2] https://www.researchgate.net/publication/393164291_Pseudo_3D_topology_optimisation_with_reinforcement_learning

[3] https://www.researchgate.net/publication/398406554_Practical_topology_optimization_with_deep_reinforcement_learning

I

• Agent A. reporting!
• Agent A., the target is active again. Proceed to Phase 2 of Operation Moonlight!
• Yes, sir!

II

Paata woke up late as always, it was past eleven o’clock. He dreamed of Her again. More than ten years had passed, and She still came to him in his dreams. His feeling had already transformed into something new, for so much time had passed, and he no longer remembered what was reality and what was a dreamed memory.

Without opening his eyes, Paata felt for his clothes and, in one fell swoop, before a match would have burned out, threw them on. And although he had never served in the Soviet army, Paata dressed rather quickly, as if by a sergeant’s call: “ Wake up and make it snappy!”

His computer, launched a week ago, was still calculating the engine’s design. This was a crucial task. The Soviet Union had no computer programs capable of automatically calculating the design of a rocket engine. Solving it would guarantee the USSR’s technological sovereignty and allow it to abandon Western technologies. And although Paata had not been an Engineer for ten years, the Design Committee still turned to him for various projects.

After brushing his teeth and grabbing a quick bite of mchadi with suluguni cheese (his favourite food, besides mamaliga), Paata looked at the terminal screen and sighed deeply:

• How much longer are you going to calculate, my friend?!
• Apparently, the equation system is pretty rigid…
• Uh, I need to call Peter to check the connections at the base station.

Paata worked as a Programmer at the Gurami Data Center, and his position allowed him to use some of the base station’s resources for his own personal tasks. One of these tasks, which Paata pursued in his free time, was Artificial Intelligence, or more specifically, its application to rocket engine design.

Personal computers were just beginning to appear in the USSR, but they had not yet reached the masses. So Paata had to connect to a Gurami base station, even though Peter had promised to send him an NC-8010 PC — the Scientific Center’s latest development — for his birthday.

Despite his young age, Paata was a sentimental and sensitive person, though he carefully concealed this from his peers. Like all key engineers in the Soviet Union, he was under the Committee’s radar, though no one ever explicitly mentioned this interest.

And now that Paata had once again taken up the task of engines design, the interest of the Committee, on the one hand, and with it that of Western players, on the other hand, had intensified.

Whether Paata was a naive dreamer or a courageous daredevil challenging the world was completely irrelevant. There were very few people in the Soviet Union willing and able to undertake such highly specialized tasks, and the Committee seized on any opportunity to explore this topic and contribute to the country’s common goal. Especially at someone else’s expense and with little efforts.

According to Department K., who oversaw the engineering sector in the Union, Paata was a rather average engineer, but his perseverance and focus on a specific task bore fruits, albeit slowly. Furthermore, Paata possessed unique experience, gained during his time as an engineer with the Soviet Space Program in the early 1970s.

In those years, both sides were in dire need of innovative solutions, as the Cold War dictated the conditions for scientific research and pushed both officials in the Ministry of Industry and academics at the Academy of Sciences. Simply twiddling their thumbs and ‘passing papers’ was no longer an option. The Party demanded progress, and the people demanded security.

It was in the early 1970s that he met Her. It was an informal gathering of literary figures in Moscow. It was not that Paata was interested in literature, but Peter, who was fond of poetry, dragged Paata to this gathering under the pretext of meeting new women.

Paata fell in love immediately, without realizing it, when he noticed emerging Her from around the corner of the lobby. Her name was Yesenia. She worked as a doctor for the international Red Cross and had come to Moscow to visit her sister, who had dragged her to this gathering under the same pretext.

• I think Anna Karenina was hysterical and could not control her emotions, which led to such a tragic end. By the way, my name is Paata, I’m a PhD student at Bauman University. I do calculations for the needs of the national economy. Nice to meet you!
• Yesenia.
• What an unusual name you have! Your parents are undoubtedly creative people to have come up with such a bright name for you.
• My father is a chemist, and my mother is a teacher. It was my older sister who gave me this name; she has been interested in literature from an early age and is full of imagination.
• Well, she was right with her choice; the name really suits the color of your eyes.
• Thank you! What does your name, Paata, mean?
• It’s a Georgian name; it means “west wind.”

Yesenia was a slim, blonde girl, short in stature, with regular, delicate features, so strikingly different from the girls Paata was accustomed to during the shifts at Baikonur Cosmodrome where he did his internship from his university.

• My friend Peter’s parents brought back the film The Godfather from their trip abroad; they’re diplomats. Would you like to see it? We’re meeting with friends at his house today!
• This is inconvenient!
• Then maybe we could take a walk in Gorky Park; I heard they are performing open-air performances of The Bremen Town Musicians there?
• No, I’m busy today.
• Could you at least tell me how to find you?
• Excuse me, I must go, goodbye!
• Goodbye!

Although Paata had not managed to find out any details about Yesenia, he was somehow certain that this would not be their last meeting.

There were also literary figures from Great Britain at the meeting, and while Paata was getting to know Yesenia, Peter was ‘building bridges internationally’.

• How do you do, Sir!
• How do you do!
• My name is Peter, I am an associate professor at Bauman Moscow State Technical University, nice to meet you, Sir!
• Nice to meet you, my name is Randy and this is my wife Clarisse, we are with the British embassy here in Moscow. And we are fervent admirers of Russian literature.
• Oh, really? My parents are diplomats; they work in Belgium.
• How interesting, a son of a diplomats, a technical guru and a lover of literature…. a very rare combination!
• Not at all, I wanted to become a philosopher in my youth and only by coincidence I became an engineer. Besides, a love for literature is instilled in Soviet schools.
• That is so true … soviet students are so advanced and open to new ideas.
• Indeed, we are!
• Cheers!

After the meeting and some casual conversations, Paata and Peter, as planned, went to watch The Godfather. Meanwhile, in a nondescript Moscow apartment in an unremarkable neighborhood, Agent A. opened his device and was preparing to receive a message from the Dolphinarium, when someone knocked on the door…

III

Clarisse, like Yesenia, was a Red Cross employee, and although they did not know each other personally, they belonged to the same organization. Clarisse’s husband, Randy, was a minor diplomat who had just been posted to Moscow and hoped this would serve as a powerful boost to his future career.

Clarisse and Randy met in Japan in the mid-1960s, at a Japanese calligraphy festival. Clarisse was interested in Japanese writing at the time (almost as much as in Russian literature), and Randy had received a scholarship from his university of foreign affairs.

Clarisse, born Ksenia, was the daughter of Russian émigrés who had emigrated west after the Russian Revolution. She was constantly searching for a safe way to visit the Soviet Union. Randy seemed like such an opportunity.

An ambitious student, a diplomat by birth, but, in fact, spoiled rich kid, Randy knew how to throw dust in people’s eyes. Although Ksenia saw through him immediately, she decided to take advantage of this opportunity to visit her ancestral homeland. Moreover, their personalities were a perfect match, which sparked their future relationship.

But now, many years later, standing in the hall of the Moscow House of Artists and Writers, Ksenia’s dream came true. She was in a country she knew only from her parents’ stories, the birthplace of Pushkin, Yesenin, and Mayakovsky. Tears welled up in her eyes…

• Why are you crying, Ksyu (as Randy affectionately called Ksenia)?
• These are tears of happiness, Bendy (Randy’s pet name).
• If I had known you would react like that, I would never have brought you here.
• No, I’m glad we came. Look at all the interesting people at this meeting!
• Yes, there are many potential candidates for our project.

Ksyu and Bendy were not afraid to communicate out loud, since they thought that no one here knew Japanese, which they spoke perfectly.

• 私たちはこのピーターに注目する必要があります。彼または彼の友人たちは私たちにとって興味深い人々である可能性があると思います。
• 私はこれに同意します

The evening gradually shifted to other venues, with people beginning to cluster around interests and seek out further banquets at local restaurants.

• Who is that girl hanging out with Peter’s friend? I think I have seen her somewhere?!
• Let me steal her from that Caucasian lion for you, my love.
• Just be careful, Bendy, we cannot arouse suspicion, you know that.

Randy intercepted Yesenia on the way out, just as she was saying goodbye to Paata.

• Excuse me, mademoiselle? Could you help me? I think I have lost my way to my wife. These Russian buildings are like labyrinths, never know where you get next.
• Sure, the main hall is over there!
• May I learn your name and occupation? I would like to get acquainted with as many people while here as possible.
• My name is Yesenia and I work for Red Cross in Berlin.
• Oh, really?! My wife works for Red Cross. She’d love to know you!
• Maybe some other time, I have to go now, bye!
• Bye-bye Yesenia, it was nice to meet you!

And even though Bendy failed to charm Yesenia, he knew that a Red Cross meeting was scheduled for the following day in Moscow, at which he hoped to continue the dialogue.

IV

Operation “Moonlight “ was a typical operation of Western intelligence agencies of the time. Its goal was to identify those among Soviet engineers and scientists sympathetic to Western values ​​and sympathetic to liberal ideas. Concepts such as empathy, sensitivity, and individuality were introduced into everyday life. According to the project’s curators, the goal was not to persuade Soviet citizens to directly cooperate with the West, but to instill in people “liberal values ​​that will help them make the right decision at the right moment,” dictated by considerations of “morality” rather than direct orders from the Soviet leadership. This influence was exerted through films, art, and literature. The end goal was to bring those scientist to the west to work for European Space Program. Specific targets — scientists and engineers of the Soviet Union — were “accompanied” according to an individual program if they “responded.”

Agent A. had joined this program at a very young age, naively, having read too many romantic spy adventures. Now he was under the watchful eye of the Dolphinarium, performing a job he disliked and which sometimes bordered on the very morals he had so fiercely defended in his youth.

Despite all this, A. was an excellent agent, the best in his class, and he performed his job impeccably. But now, when there was a knock on his door, perhaps for the first time in his life, he realized he might have been blown.

V

Agent A. calmly stood up, closed the lid of the music box where he hid the transmitter, quickly spread the magazines on the floor, and walked to the door.

Looking through the peephole, he saw his neighbor, Marinka.

Agent A. reluctantly opened the door.

• Hi Sasha! The Bremen Town Musicians are showing in Gorky Park! Shall we go!?
• Right now?
• Yes!
• Oh, Marinka, I’m just knitting a new sweater for Nastya. Look, cannot you see the magazines lying on the floor?!

Nastya was Agent A.’s five-year-old daughter.

• Oh, come on, Sasha, you can finish knitting later!

Agent A. reluctantly threw on his shawl and went with Marinka to Gorky Park. There were two reasons why Agent A. so quickly and easily missed the important transmission from the Dolphinarium: first, the transmitter had an automatic message recording function; and second, and more importantly, Agent A. wasn’t certain that Marinka wasn’t working for the Committee, even though he had known her for seven years.

Marinka was a typical girl of easy morals. A fashionista and a spendthrift, she plied wealthy men and spent all their money. But from time to time, she had a tendency to disappear for several days, which alarmed Agent A.

Her latest “victim” was a certain Georgian or Armenian. A talented, promising young engineer whom Marinka met at a friend’s wedding. His name was either Lasha or Pasha, and according to Marinka, he was expecting to be assigned to Moscow after completing his studies. He already had a temporary job. Consequently, he had the money he so generously spent on Marinka. She had him so befuddled that she was fleecing him. Even though they had not been intimate yet. Either Lasha or Pasha was constantly giving Marinka gifts and taking her to expensive restaurants.

• By the way, how’s your Armenian?
• He’s Georgian! And tonight, he’s taking me to the Kura restaurant. Want to come with us?! He’s so smitten with me that he will pay for both of us if I ask him.
• Oh, Marinka, you’re always on your best behavior!

Even though Agent A. laughed it off, he seriously considered going on a date with this Georgian guy with Marinka that night. Young, talented Soviet engineers were exactly the kind of contacts Agent A. needed for the new Dolphinarium project.

VI

Mission Moonlight — a European project to establish a dedicated satellite constellation orbiting the Moon.

The primary goal for 1972 is to provide advanced communication and navigation services for the rapidly increasing number of lunar missions planned by various countries over the next few decades.

VII

Directive 70–75–73–73–79

From European Space Consortium’s meeting

January 1, 1973

To ensure successful completion of project Moonlight, ESC decided to broaden search for advance technologies especially those related to automated design and manufacturing. To guarantee success, it is acceptable to ‘borrow’ technologies from Warsaw Pact including but not limited to key personnel.

All agencies need to join efforts in order to bring talented scientist to the West and organize proper work conditions.

I instruct all relevant services and agencies to approach this issue creatively. The use of any methods and tools, even the most extreme, is permitted.

Director of Dolphinarium: C. Jarie

VIII

Agent Morozko’s Report

January 2, 1973

The fish are splashing around in the Dolphinarium again. The operation has been given its final name, Operation “Moonlight “. I’m attaching the text of the meeting.

IX

Having returned from the park and finally read the message about the new directive from the Dolphinarium, Agent A. decided to join Marinka on her date with the Georgian that evening.

The Kura Restaurant was a typical Georgian restaurant with authentic Georgian cuisine, a welcoming atmosphere, and an ethnic interior.

Paata, Marinka’s suitor, chose this restaurant not by chance, but because he knew a chef who worked there and gave a good discount for a kickback.

Seeing Marinka and Sasha, Paata was a little surprised, but he didn’t show it.

• This is my friend Sasha, she’s a single mother, and I sometimes take her out.

Marinka was as simple as a pie.

• Nice to meet you! said Paata.

A warm burning sensation ran through Sasha’s body, and a barely noticeable blush appeared on her cheeks. The Dolphinarium’s regulations forbade romantic relationships with subjects, but ultimately, Sasha (Agent A.) was just a person, a young woman not averse to affection.

After a formal introduction, Paata placed an order:

• Khinkali, chacha, and beer, please!

Paata was either hoping to lift Marinka’s spirits, or he had completely given up hope and decided to at least get drunk.

• And I’ll have Abrau-Durso champagne! And what is for you, Sasha?

As expected, Marinka did not spare Paata’s money and ordered everything she could.

• Just a salad for me.

Sasha said modestly.

She immediately recognised Paata’s nature, who loved fallen women, but respected decency and innocence. And so, she decided to feign modesty.

Sowing the seed of doubt and allowing it to grow on its own — that was Agent A.’s main task that evening. Spark an idea and let it take on a life of its own.

• Have you read The Gulag Archipelago, Paata?
• No, what is it, Sasha?
• It’s a story by the talented Russian writer Solzhenitsyn, and its publication is currently being prepared in Paris.
• I don’t like fiction.
• What do you like?
• Marx and Engels

Paata hated Marx and Engels, as well as all this Scientific Communism. They bored him, but he was seeing Sasha for the first time in his life and decided to test the waters first.

• Marx and Engels?! How boring.

Sasha burst out laughing.

• Well, I didn’t expect any other answer, from such an ardent supporter of Soviet ideology.

Although Paata didn’t say anything about it, for some reason he created that impression in all his new acquaintances.

• Are you Alexandra Vnukova?

The waiter, who suddenly approached, asked Sasha.

• Yes!
• You are wanted on the phone!

X

• Hello!
• Sasha, this is Uglya Ivanovna!
• Hello, Uglya Ivanovna, how is Nastya? Is she behaving well? Thank you for agreeing to babysit her. Did something happen?
• Yes, the neighbors found a dead cat at your door and called the sanitary and epidemiological station. They suspect distemper and ask everyone to come to the clinic tomorrow for questioning.
• Is it dangerous for humans?
• No! But protocol requires everyone to be questioned.
• Thank you for letting me know, Uglya Ivanovna. Anything else?
• Yes! Nastya was very scared!
• Thank you, I’ll be home soon.

Unexpectedly for Paata and Marinka, Sasha said goodbye and went home to reassure Nastya.

She got into a taxi conveniently parked near the restaurant and sped home. Somewhere halfway through the journey, she realized the driver was drunk, but she decided not to get out of the taxi and instead drove to the end of the route.

That evening, Paata, as always, returned home empty-handed. Sasha, meanwhile, was plagued by vague thoughts.

• First the dead cat, and then the drunk driver — is someone giving me some kind of a signal?! Should I be worried? Should I call the Dolphinarium?!

Sasha thought as she settled into her warm bed.

XI

The day after the literary gathering, a meeting of the Moscow Red Cross office was held on Bolshaya Lubyanka.

Among the others present were Ksyu (as a member of the organization), Bendy (as a major sponsor), and Yesenia.

On the agenda was the opening of a new office in China. The council was recruiting volunteers for a year-long assignment to the Middle Kingdom.

Ksyu, knowing the agenda in advance, had already agreed to a year-long separation from Bendy. Frankly, their relationship had not been going well lately, and the opportunity to take a break from each other for a while seemed like a good idea.

Yesenia, who had recently broken up with her abusive boyfriend and was looking for an opportunity to hide her feelings on a trip abroad, also expressed a desire to go to China.

• Mademoiselle Yesenia!
• Hello!
• We saw each other yesterday at the literary gathering. My name is Randy, and this is my wife, Clarisse. Nice to meet you again! We were just discussing this wonderful opportunity to visit a new country and imbibe Asian culture. I see you’re planning on going too?!
• Yes, China is in dire need of humanitarian aid due to the imminent end of the Cultural Revolution.
• It’s very commendable that you care so much about a people so foreign to you!
• I’m just doing my job.
• I think you should befriend Clarisse; it will make you safer during your trip, and I’ll feel calm about my wife. How about I arrange for the Chairman to put you in the same room?!

And despite Yesenia being a cautious girl, she immediately took a liking to Ksyu. Her inner melancholy was similar to the feelings Yesenia was experiencing herself.

Yesenia agreed to Randy’s proposal.

At the end of February, the girls joined the group and set off on the trip.

XII

After defending his PhD thesis and receiving, as expected, a job assignment to Moscow, Paata found work at the Moscow office of Zenith, a large industrial holding company engaged in space exploration.

The Moscow office was Zenith’s central headquarters; all major space operations were planned and calculated here.

In just one year, Paata rose to a manager. This was facilitated by his initiative to create internal regulations for conducting engineering and technical calculations.

Everything in Paata’s life was going well, but he could not forget that meeting with Yesenia at a literary gathering.

Through Peter, who had been in contact with Randy since that meeting, he learned that Yesenia and Ksyu were on a trip together in China, and he decided to pass on a letter to Yesenia. It is not known exactly what Paata wrote to Yesenia in his first letter, but they slowly began communicating by correspondence, apparently because Yesenia wanted to distract herself from her memories of her abusive ex-boyfriend.

One word led to another, Paata declared his love to Yesenia and was already preparing to go to China on a friendly visit.

In conversations with Ksyu, Yesenia said she really admired Paata’s courtship skills. This is probably why, when Paata briefly visited China, he and Yesenia fell in love.

From then on, Paata and Yesenia became best friends, soulmates, and a loving couple. They spent hours talking on the phone — at Peter’s house, his parents had a special diplomatic telephone line with various countries. Yesenia visited Paata in Moscow many times, and they also traveled together to other Asian countries, such as Vietnam, which had just won the war against the Americans.

In Vietnam, Paata and Yesenia met a young student, Jasmine, who had won numerous math competitions and showed great promise for the Communist Party of Vietnam. Jasmine was very young and beautiful, a kind of Asian flower, glowing in the rays of Saigon. Paata even briefly considered picking this flower, but his love for Yesenia kept him from such foolishness.

XIII

Everything in our heroes’ lives was going well, if only not for the activities of the Dolphinarium.

Through agents of influence like A., the Dolphinarium promoted Western empathic values, and since there was nothing to counteract this “good,” this bore fruit.

In the case of our hero Paata, agent A., who was tasked with supervising him, decided to play the Georgian card.

In those years, an underground movement of people dissatisfied with the infringement of national identity and the status of the Georgian language was emerging in Georgia. As a true Georgian patriot, Paata also fell under the influence of this sentiment. And since Paata was a simple and kind-hearted person, this sentiment also influenced his work.

Zenith, in essence, was an arm of the Party’s interests, which in turn was pursuing a harsh policy in the Caucasus. Paata understood this, and his nerves were slowly fraying.

The culmination of these experiences occurred on March 12, 1975, when Paata was hospitalized at the Neurology Institute with a nervous breakdown. Doctors banned him from work for a month and prescribed him hospitalization.

The Committee knew that Paata was being under development by the Dolphinarium through its agent Morozko, and when Paata was hospitalized, they decided to “test” him to see how deeply his sentiments penetrated. The Committee’s ultimate goal was to “reorient” Paata’s mind toward correct Soviet values ​​while he was at the peak of his nervous breakdown with his brain still malleable and susceptible to external influences.

But the Dolphinarium was not idle either.

XIV

There were three people in the ward: Patient 1, Patient 2, and Paata.

Paata lay in a comfortable corner on his bed, half reading a book about chess, half listening to the conversation between his two fellow sufferers.

• Oh, our team lost the football game again!
• It would have been better if they hadn’t played at all. Dynamo has gotten fat lately, but Spartak is a whole other story! New training methods, new equipment, a new coach finally — everything is modern and brilliant.
• I’ve been a Dynamo fan all my life, and my father was a Dynamo fan, and my grandfather was too!
• What’s this Dynamo, this Lobanovskyi of yours? He’s not a one, he’s a complete zero!
• We’ll see; the season is still a long way off.
• What’s there to see? And what do you think, Paata? Which is better: Dynamo’s old and reliable approach or Spartak’s modern methods?

For some reason, at that moment, Paata felt uneasy, and a strange thought entered his head.

He excused himself and went to the restroom to get some water. He saw the same scene in the hallway. For some reason, everyone (patients and staff) was divided into couples, or at most triples. Paata felt as if they were all arguing about something, and when two were talking, the third would remain silent and listen.

Paata suddenly remembered a spy novel he’d read as a child. In it, every Soviet agent had a Western counterpart, who constantly kept tabs on each other and were never alone. Sometimes a third “counterpart” from Asia would join them, but he or she was always silent.

Paata quickly drank some water, returned to the ward, lay down on his side, and fell asleep.

A month later, he was discharged. Yesenia didn’t come to meet him. Only his mother, who had come for that from Tbilisi, met Paata.

XV

Dynamo won gold that year…

XVI

After his discharge, due to mental and nervous exhaustion, Paata was no longer able to fully perform his job and decided to resign in early April. Yesenia did not support this decision.

Paata’s brother, Esarion, advised him to consider the new profession of programmer, which was just beginning to emerge in the USSR. Although the profession was new to the Union, Esarion had extensive experience in this specialty, having worked at the Farhad defense plant (the Soviet branch of the Ministry of Defense in the Middle East).

A week after his discharge, Paata went to visit his friend Peter. They had the following conversation:

• So, how’s your father, Peter? On a business trip again?
• Yes, he recently left; he’s been assigned to Zurich.
• Very interesting, and what does he tell you about Switzerland?
• Lots of things, for example, that story about sleeping agents.
• Sleeping agents? What is that? Lazy people?
• No, rumour has it this is a new Western program, where a person doesn’t know they’re an agent, but subconsciously carries out actions aimed at undermining our sovereignty.
• What do you mean, subconsciously? Are they hypnotized?
• That’s all I know from my father.
• Don’t worry! If the West has such programs, then we probably have them too. Right?
• Plight. Fight.
• Blight.

The friends laughed.

XVII

Yesenia unexpectedly leaves Paata without explanation and extends her business trip to China.

She and Paata never saw each other again.

XVIII

In 1978, protests against attempts to infringe on the rights of the Georgian language broke out in Georgia. Following this, Paata decided to move to Tbilisi, driven by his national pride.

Peter went with him.

They found jobs at Gurami, the leading computing center of the Georgian SSR, having already worked as programmers for several years.

That same year, Agent A. died mysteriously in Moscow.

Nastya was adopted by Marinka (and perhaps by the Soviet government).

XIV

Paata woke up late as always, it was past eleven o’clock…

I

• Агент А. на связи!
• Агент А., объект снова активен, приступайте ко второй фазе операции Плавленый сырок!
• Слушаюсь!

II

Паата проснулся как всегда поздно, уже был одиннадцатый час. Ему снова снилась Она. Прошло уже более десяти лет, а Она до сих пор приходила к нему во снах. Его чувства уже трансформировалась в нечто новое, ведь прошло так много времени, и он уже не помнил, где явь, а где приснившееся воспоминание.

Не открывая глаз, Паата нащупал одежду и одним махом, за время пока догорела бы спичка, накинул ее на себя. И хотя он никогда не служил в советской армии, Паата одевался довольно быстро как по команде — “Рота подъем!“.

Его ЭВМ, запущенный неделю назад, до сих пор вычислял задачу морской трубоукладки. Это была очень важная задача. Программ способных правильно рассчитать динамику укладки трубы с судна в Советском Союзе еще не было. Ее решение гарантировало бы технологический суверенитет СССР и позволило бы отказаться от западных технологий. И хотя Паата уже как десять лет не был Инженером, Конструкторский Комитет до сих пор обращался к нему за расчетами для разных проектов.

Почистив зубы и быстро перекусив мчади с сулугуни (его любимая еда, не считая мамалыги), Паата посмотрел на экран терминала и глубоко вздохнул:

• Ну сколько ты еще будешь считать, дружок?!
• Видимо система уравнений жесткая…
• Эх, надо позвонить Петру, чтобы проверил соединения на базовой станции.

Паата работал Программистом в вычислительном центре ‘Гурами’ и его положение позволяло ему использовать часть ресурсов базовой станции для ‘собственных задач’. Одной из таких задач была трубоукладка, но другое увлечение Пааты, которым он также занимался в свободное время, был Искусственный Интеллект, а точнее его приложение к проектированию ракетных двигателей.

Персональные компьютеры еще только начинали появляться в СССР, но до широких масс они еще пока не доходили. Поэтому Паате приходилось подключаться к базовой станции ‘Гурами’, хотя Петр обещал ‘подогнать’ ПК НЦ-8010 к его дню рождения — новейшую разработку Научного центра.

Паата несмотря на его не пожилой возраст был человеком сентиментальным и переживательным, хотя тщательно это скрывал среди своих сверстников. Так же, как и все ключевые инженеры в Советском Союзе он находился в поле зрения Конторы, хотя явно об этом интересе к нему ему никто никогда не говорил.

И сейчас, когда Паата снова занялся задачей трубоукладки, активизировался и интерес Конторы, а вместе с ней и западных игроков.

Был ли Паата наивным мечтателем, или отважным смельчаком, бросившим вызов миру, было совершенно не важно. Желающих и способных заниматься такими узкоспециализированными задачами в Союзе было очень мало, и Контора цеплялась за любую возможность исследовать эту тему и внести вклад в общее дело страны. Особенно за чужой счет и малыми усилиями.

По оценке Отдела К., курировавшего в Конторе инженерную сферу в Союзе, Паата был довольно средним инженером, но его усидчивость и фокусировка на конкретной задаче приносили свои плоды, хотя и не очень быстро. Кроме того, Паата обладал уникальным опытом, полученным им во времена работы инженером на морском строительном судне ‘Удача’ на проекте ‘Шелест’ в Охотском море в начале 70-х.

В те годы обе стороны особенно остро нуждались в инновационных решениях так как каноны Холодной Войны диктовали свои условия ведения научных исследований и подгоняли как чиновников в министерстве промышленности, так и академиков в Академии Наук. Просто протирать штаны и ‘перебрасываться бумажками’ уже не получалось. Партия требовала прогресса, народ требовал безопасности.

Именно в начале 70-х он и встретил Её. Это было неформальное собрание литераторов в Москве. Не то, чтобы Паата интересовался литературой, но Петр, увлекавшийся поэзией, затащил Паату на это собрание под предлогом новых знакомств с прекрасным полом.

Паата влюбился сразу, сам этого не поняв, он обратил внимание на скромную девушку выплывшую из-за угла вестибюля. Ее звали Есения. Она работала врачом в международной организации Красный Крест и приехала в Москву к сестре, которая и затащила ее на это собрание под тем же предлогом.

• Мне кажется, что Анна Каренина была истеричкой, и не умела управлять своими эмоциями, что и привело ее к такому печальному концу. Кстати, меня зовут Паата, я аспирант Бауманского Университета и занимаюсь вычислениями для нужд народного хозяйства. Приятно познакомиться.
• Есения.
• Какое у вас необычное имя, ваши родители несомненно творческие личности что придумали вам такое яркое имя.
• Мой отец химик, а мама учительница, это моя старшая сестра дала мне такое имя, она с ранних лет увлекается литературой и на выдумку полна.
• Что ж, она угадала с выбором, имя очень идет к цвету ваших глаз.
• Спасибо! А что значит ваше имя — Паата?
• Это грузинское имя, оно значит “западный ветер”

Есения была белокурая стройная девушка, невысокого роста, с правильными нежными чертами лица и так разительно отличалась от тех девушек, к которым привык Паата на вахтах в море, на которых он проходил практику от своего университета.

• Родители привезли моему другу Петру, фильм Крестный Отец, из своей заграничной поездки, они дипломаты, не хотите ли посмотреть? Мы сегодня собираемся с друзьями у него дома!
• Это неудобно!
• Тогда может прогуляемся в парке Горького, я слышал там идут спектакли Бременские Музыканты под открытым небом?
• Нет, я сегодня занята.
• Скажите хотя бы как вас найти?
• Извините мне надо идти, до свидания!
• До свидания!

И хоть Паате не удалось разузнать подробности о Есении, почему-то он был уверен, что это была их не последняя встреча.

Были на собрании, и литераторы из Великобритании, и тем временем пока Паата знакомился с Есенией, Петр наводил мосты на международном уровне.

• How do you do, Sir!
• How do you do!
• My name is Peter, I am an associate professor at Bauman Moscow State Technical University, nice to meet you, Sir!
• Nice to meet you, my name is Randy and this is my wife Clarisse, we are with the British embassy here in Moscow. And we are fervent admirers of Russian literature.
• Oh, really? My parents are diplomates; they work in Belgium.
• How interesting, a son of a diplomates, a technical guru and a lover of literature…. a very rare combination!
• Not at all, I wanted to become a philosopher in my youth and only by coincidence I became an engineer. Besides, a love for literature is instilled in Soviet schools.
• That is so true … soviet students are so advanced and open to new ideas.
• Indeed, we are!
• Cheers!

После собрания и случайных диалогов, Паата и Петр, как и планировали отправились смотреть Крестного Отца. А тем временем в одной невзрачной Московской квартире в ничем не примечательном районе, агент А. открыл свое устройство и готовился принять сообщение из Дельфинария. И тут в дверь постучали…

III

Clarisse, так же как и Есения, была сотрудником Красного Креста и хотя друг друга они лично не знали, они принадлежали к одной и той же организации. Муж Clarisse, Randy был мелким дипломатом, только что получившим назначение в Москву и рассчитывающим, что это назначение послужит мощным импульсом в его дальнейшей карьере.

Clarisse и Randy познакомились в Японии в середине 60-х, на фестивале японской каллиграфии. Clarisse в то время интересовалась японской письменностью (почти также сильно как и Русской литературой), а Randy получил путевку от своего университета иностранных дел.

Clarisse, по рождению Ксения, была дочкой Русских эмигрантов переселившихся на запад после русской революции. И всеми путями искала возможность безопасно посетить Советский Союз. Randy казался ей тогда такой возможностью.

Честолюбивый студент, потомственный дипломат, а по сути, сынок богатых родителей, Randy умел бросать пыль в глаза. Хотя Ксения его сразу раскусила, она решила использовать эту возможность посетить историческую родину. Кроме того, они идеально подходили характерами, что и явилось стимулом для их дальнейших отношений.

Теперь много лет спустя, находясь в зале Московского Дома Художников и Литераторов, мечта Ксении исполнилась. Она находилась в стране, о которой знала только из рассказов родителей, родине Пушкина, Есенина и Маяковского. Слезы наворачивались на ее глаза…

• Почему ты плачешь Ксю (так Randy ласково называл Ксению)?
• Это слезы счастья Бенди (домашнее прозвище Randy )
• Знал бы я что ты так отреагируешь ни за что не привел бы тебя сюда
• Да нет я рада что мы пришли, посмотри какие интересные люди на этом собрании!
• Да много потенциальных кандидатов для нашего проекта

Ксю и Бенди не опасались общаться вслух, так как думали что японского, на котором они говорили в совершенстве, тут никто не знал.

• 私たちはこのピーターに注目する必要があります。彼または彼の友人たちは私たちにとって興味深い人々である可能性があると思います。
• 私はこれに同意します

Вечер плавно перетекал в другие места, народ начинал кучковаться по интересам и искать продолжений банкета в местных ресторанах.

• Кто это девушка, которая общается с этим другом Петра? Мне кажется я ее где-то видела?!
• Позволь мне украсть ее из рук этого кавказского льва для тебя любимая.
• Хорошо только будь осторожен Бенди, мы не можем вызывать подозрения ты же знаешь.

Randy перехватил Есению на выходе, как раз в тот момент, когда она распрощалась с Паатой.

• Excuse me, mademoiselle? Could you help me? I think I have lost my way to my wife. These Russian buildings are like labyrinths, never know where you get next.
• Sure, the main hall is over there!
• May I learn your name and occupation? I would like to get acquainted with as many people while here as possible.
• My name is Yesenia and I work for Red Cross in Berlin.
• Oh, really?! My wife works for Red Cross. She’d love to know you!
• Maybe some other time, I have to go now, bye!
• Bye-bye Yesenia, it was nice to meet you!

И несмотря на то, что Бенди не удалось обаять Есению, он знал что на завтра в Москве было назначено собрание Красного Креста на котором он рассчитывал продолжить диалог.

IV

Операция ‘Плавленый сырок’ была типичной операцией западных спецслужб того времени. Целью ее было выявление в рядах советских инженеров и ученых сочувствующих ценностям запада и сопереживающих либеральным идеям. В обиход вводились такие понятия как эмпатия, чувственность и индивидуальность. По задумке кураторов проекта суть была не в том, чтобы склонить советских граждан к прямому сотрудничеству с западом, а в том, чтобы привить людям “либеральные ценности, которые в нужный момент помогут им принять правильное решение” продиктованные соображениями “морали”, а не прямыми указаниями Советского начальства. Воздействие оказывалось через кино, искусство и литературу. И не имело какой-то сформулированной конечной цели. Конкретные объекты — ученые и инженеры советского союза — “сопровождались” по индивидуальной программе в случае появления “отклика” с их стороны.

Агент А. присоединился к этой программе в очень раннем возрасте по наивности души, начитавшийся романтических книг о приключениях шпионов, теперь же он находился под пристальным взором Дельфинария и выполнял работу, которая ему совсем не нравилась и порой граничила с той самой моралью, которую он в юности так яростно защищал.

Несмотря на все это, А. был отличным агентом, лучшим в своем выпуске, и выполнял свою работу безупречно. Но сейчас, когда в его дверь постучали, возможно он впервые в жизни задумался о том, что засыпался.

V

Агент А. спокойно встал, закрыл крышку музыкальной шкатулки, где он прятал передатчик, быстро разложил журналы на полу и подошёл к двери.

Посмотрев в глазок, он увидел соседку Маринку.

Агент А. нехотя открыл.

• Привет Саша! В Парке Горького показывают спектакли Бременских музыкантов! Пойдем!?
• Сейчас?
• Да!
• Ой Маринка, я как раз вяжу новую кофту Насте, вон, не видишь журналы валяются на полу?!

Настя была пятилетней дочерью агента А.

• Ой, да пошли Сашка, потом довяжешь!

Агент А. нехотя накинул шаль и поехал вместе с Маринкой в Парк Горького. Причин по которой агент А. столь быстро и легко решил пропустить важную передачу из Дельфинария было две: во-первых, передатчик обладал функцией автоматического записывания сообщения, а, во-вторых, что более важно, Агент А. не был уверен, что Маринка не работает на Контору, хоть знал ее уже целых семь лет.

Маринка была типичной профурсеткой. Модница, транжира, она садилась на шею состоятельных мужчин и тратила все их деньги. Но время от времени, имела свойство пропадать на несколько дней, что и настораживало Агента А.

Ее последней “жертвой” был некий то ли грузин, то ли армянин. Молодой талантливый, подающий надежды инженер с которым Маринка познакомилась на свадьбе у друзей. Его звали то ли Лаша, то ли Паша и, по словам Маринки, он ожидал распределения в Москву после окончания обучения. У него уже была временная работа. А следовательно были и деньги, которые он столь щедро тратил на Маринку.

Она настолько запудрила ему голову, что умудрялась обдирать его как липку. Хотя близости между ними еще не было. То ли Лаша, то ли Паша постоянно дарил Маринке разные подарки и водил по дорогим ресторанам.

• Кстати, как там твой Армянин?
• Он Грузин! И сегодня вечером он ведет меня в ресторан Кура. Хочешь пошли с нами?! Он так увлечен мной что заплатит за нас двоих если я его попрошу.
• Ой, Маринка, ты все в своем репертуаре!

Несмотря на то что агент А. отшутился, он серьезно задумался о том не пойти ли сегодня вместе с Маринкой на свидание с этим Грузином. Молодые талантливые советские инженеры были тем типом знакомств которые нужны были агенту А. для нового проекта Дельфинария.

VI

Директива организации Дельфинарий # d185d183d0b9.

Операция Разжижение Мозгов (рабочее название).

Заседание от 30 Декабря, 1972 года.

В связи с усилением Союзов, и снижением возможности прямой конкуренции, предписываю применить альтернативный, творческий подход для решения вопроса обеспечения доминирования демократических сил в современном миропорядке. В свете невозможности получения устойчивого технологического преимущества, правительство и Президент вместе с союзниками решили зайти с другой стороны. А именно противопоставить идеалы инженерного общества СССР, эмпатической морали запада, склонив ключевых инженеров и ученых Союза колебаться и сомневаться при принятии значимых решений.

Предписываю всем соответствующим службам и органам подойти к решению данного вопроса творчески. Допускается использование любых методов и инструментов, даже самых крайних.

Директор Дельфинария - В.П.К.

VII

Донесение Агента Морозко

2 Января 1973 года

В Дельфинарии снова плещется рыбка. Операция получила окончательное название “Плавленый сырок”. Текст с заседания прилагаю.

VIII

Вернувшись из парка, и прочитав наконец сообщение о новой директиве из Дельфинария, агент А. решил присоединиться к Маринке на этом свидании с Грузином сегодня вечером.

Ресторан Кура, был типичным грузинским рестораном с аутентичной грузинской кухней, радушной атмосферой и этническим интерьером.

Паата, так звали ухажера Маринки, выбрал этот ресторан не случайно, а в связи с тем что у него тут работал знакомый повар делающий хорошую скидку за откат.

Увидев Маринку и Сашу, Паата немного удивился, но виду не подал.

• Это моя подруга Саша, она мать одиночка и я иногда вывожу ее в свет.

Маринка была проста как три копейки.

• Приятно познакомиться!

Сказал Паата.

Теплое жжение пробежало по телу Саши и на щеках появился еле заметный румянец. Предписания Дельфинария запрещали романтических отношений с объектами, но в конце концов Саша была всего лишь человеком, молодой женщиной не чуждой ласке.

После формального знакомства, Паата сделал заказ:

• Хинкали, чачу и пиво, пожалуйста!

Паата то ли надеялся поднять настроение Маринке до нужного градуса, то ли уже просто отчаялся окончательно и решил просто хотя бы напиться.

• А мне шампанское Абрау-Дюрсо! А тебе что Саша?

Как и предполагалось Маринка не жалела денег Пааты и заказывала на всю катушку.

• Мне просто салат.

Скромно сказала Саша.

Она сразу раскусила натуру Пааты. Любящего падких женщин, но уважающего порядочность и невинность. И решила притвориться скромницей.

Посеять зерно сомнения и позволить ему самому прорасти, вот что было основной задачей агента А. на этот вечер. Зародить идею и уйти в сторону, позволив ей начать жить собственной жизнью.

• Вы читали Архипелаг Гулаг, Паата?
• Нет, а что это, Саша?
• Это рассказ талантливого русского писателя Солженицына, его издание готовится сейчас в Париже.
• Не люблю беллетристику.
• А что же вы любите?
• Маркса и Энгельса

Паата, ненавидел Маркса и Энгельса, а так же весь этот Научный коммунизм. Они нагоняли на него тоску, но Сашу он видел впервые в жизни и решил сначала прощупать почву.

• Маркса и Энгельса?! Вот это скука.

Расхохоталась Саша.

• А, впрочем, я другого ответа и не ожидала, от ярого сторонника советской идеологии.

Хоть Паата ничего об этом не говорил, но он почему то создавал такое впечатление у всех своих новых знакомых.

• Вы Александра Внукова?

Спросил неожиданно подошедший официант, Сашу.

• Да!
• Вас к телефону!

IX

• Алло!
• Саша, это Аглия Ивановна!
• Здравствуйте Аглия Ивановна, как там Настя? Хорошо себя ведет? Спасибо что согласились с ней посидеть. Что-нибудь случилось?
• Да, соседи обнаружили мертвую кошку у вашей двери в подъезде и вызвали санэпидемстанцию. Они подозревают чумку, и просят завтра всех прийти в диспансер для опроса.
• Это опасно для человека?
• Нет! Но протокол требует всех опросить.
• Спасибо, что сообщили Аглия Ивановна, что-нибудь еще?
• Да! Настя очень сильно испугалась!
• Спасибо, я скоро буду.

Саша неожиданно для Пааты и Маринки попрощалась и отправилась домой, успокаивать Настю.

Она села в удачно стоявшее около ресторана такси и с ветерком помчалась домой.

Где-то на середине пути она поняла что водитель был не трезв, но решила не выходить из такси, а доехать до конца дороги.

Тем вечером, Паата, как всегда ни с чем вернулся домой. Сашу же терзали смутные мысли.

• Сначала мертвая кошка, а потом и пьяный водитель, кто-то подает мне какие-то сигналы?! Следует ли мне волноваться? Следует ли сообщить в Дельфинарий?!

Думала Саша укладываясь в свою теплую постель.

X

На следующий день после собрания литераторов, на Большой Лубянке состоялось заседание московского офиса Красного Креста.

На нем присутствовали по мимо всех прочих: Ксю (как член организации), Бенди (как крупный спонсор), также пришла и Есения.

На повестке дня было открытие нового офиса в Китае. Совет набирал желающих для годовой командировки в Поднебесную.

Ксю зная повестку дня заранее, уже согласовало годовую разлуку с Бенди. Честно говоря, в последнее время их отношения не ладились и возможность отдохнуть друг от друга и перезагрузить чувства казались им хорошей идеей.

Так же желание поехать в Китай высказала и Есения, которая недавно рассталась со своим парнем моральным-абьюзером, и искала возможности спрятать свои чувства в заграничной поездке.

• Мадмуазель Есения!
• Здравствуйте!
• Мы виделись вчера на собрании литераторов, меня зовут Randy, а это моя жена Clarisse. Nice to meet you, again! Мы как раз обсуждали эту прекрасную возможность посетить новую страну и впитать культуру Азии. Я смотрю, Вы тоже собираетесь?!
• Да, Китай остро нуждается в гуманитарной помощи в связи с неизбежным окончанием Культурной Революции.
• Это очень похвально, что Вы так заботитесь о чуждом для Вас народе!
• Я просто делаю свое дело.
• Я считаю, что вам нужно подружиться с Clarisse, так и Вам будет безопаснее во время поездки и мне спокойнее за свою жену. Давайте я договорюсь с Председателем чтобы вас вместе расквартировали?!

И несмотря на то, что Есения была девушка осторожная, Ксю ей почему-то сразу понравилась. Ее внутренняя тоска была похожа, на те чувства, которые испытывала Есения сама.

Есения согласилась на предложение Randy.

В конце февраля девушки присоединились к группе и отправились в командировку.

XI

Защитив кандидатскую диссертацию и получив, как и ожидалось, распределение в Москву, Паата устроился в Московский офис предприятия Зенит — крупный промышленный холдинг, занимающийся морским строительством на крайнем Севере.

Московский офис был так называемым мозговым центром Зенита, все крупные морские операции планировались и рассчитывались тут.

Всего за год Паата вырос до руководителя. Этому способствовало его инициатива по созданию внутреннего регламента проведения инженерно-технических расчетов.

Все в жизни Пааты складывалось удачно, но он не мог забыть той встречи с Есенией на собрании литераторов.

Через Петра, который с той самой встречи общался с Randy, он узнал, что Есения и Ксю вместе находятся в командировке в Китае, и решился передать письмо Есении.

Доподлинно не известно, что Паата написал Есении в своем первом письме, но потихоньку-полегоньку они начали общаться по переписке, видимо потому, что Есения хотела отвлечься от своих воспоминаний о бывшем парне-абьюзере.

Слово за слово, сковородой по столу, Паата признался в любви Есении и уже собирался в Китай с дружеским визитом.

В разговорах с Ксю, Есения говорила, что ей очень нравится, как Паата умеет ухаживать. Видимо поэтому, когда Паата ненадолго приехал в Китай у них с Есенией случилась настоящая чистая любовь.

С тех пор Паата и Есения стали лучшими друзьями, родственными душами и влюбленной парой. Они часами говорили по телефону- в доме Петра, у его родителей была специальная дипломатическая телефонная связь с разными странами.

Есения много раз приезжала в Москву в гости к Паате, так же они осуществляли совместные поездки по другим странам Азии, таким как Вьетнам, например, которая только что выиграла войну с Американцами.

Во Вьетнаме Паата и Есения познакомились с молодой студенткой Жасмин, которая была победительницей многих математических олимпиад и подавала большие надежды для КПВ. Жасмин была очень юна и красива. Этакий азиатский цветок, пылающий в лучах Сайгона. Паата даже краем сознания задумывался, а не сорвать ли этот цветок, но любовь к Есении удерживала его от подобных глупостей.

XII

Все в жизни наших героев шло хорошо, если бы только не активность Дельфинария.

Через агентов влияния подобных А., Дельфинарий занимался пропагандой западных эмпатических ценностей, и потому, что ‘добру’ противопоставить было нечего, это приносило свои плоды.

В случае с нашим героем Паатой, агент А., которой было поручено его курирование, решила разыграть Грузинскую карту.

В те годы в Грузии прорисовывалось подпольное движение недовольных ущемлением национальной идентичности и статусом грузинского языка. Под влияние сентимента, как истинный патриот Грузии попал и Паата. И так как Паата был простым и добрым в душе человеком, этот сентимент оказывал и влияние на его работу.

Зенит, по сути, был щупальцем интересов Партии, которая в свою очередь проводила жесткую политику на Кавказе. Паата это понимал и потихоньку его нервы расшатывались.

Кульминация этих переживаний произошла 12 Марта 1975 года, когда Паату госпитализировали в институт неврологии с нервным срывом. Врачи на месяц запретили ему работать и прописали стационарный режим.

Контора знала, что Паата находился в разработке Дельфинария через своего агента Морозко, и когда Паата попал в больницу, решила его “ прощупать” на глубину проникновения сентимента. Конечной целью Конторы было “переориентировать” мозг Пааты на правильные советские ценности пока он находился в точке максимума нервного срыва и его мозг был пластичен и подвержен влиянию из вне.

Но и Дельфинарий не сидел без дела.

XIII

В палате было трое: Пациент 1, Пациент 2 и Паата.

Паата лежал в комфортном углу на своей койке и наполовину читал книгу про шахматы, на половину слушал разговор двух друзей по несчастью.

• Эх, наши опять в футбол проиграли!
• Да лучше бы вообще не играли, это Динамо в последнее время обросло жирком, то ли дело — Спартак! Новые методики тренировок, новое оборудование, новый тренер наконец — все современное и блестящее.
• Я всю жизнь болел за Динамо, и мой отец болел за Динамо, и мой дед тоже!
• Да, что это Динамо, ваш Лобановский, он не единица, он полный ноль!
• Поживем увидим, до конца сезона еще далеко.
• Да, что там смотреть то, а Вы как думаете Паата? Что лучше, старый и надежный подход Динамо или современные методики Спартака?

Почему-то в этот момент Паате стало не по себе, странная мысль зародилась у него в голове.

Он извинился и вышел в уборную набрать воды. И увидел в коридоре ту же самую картину. Почему-то все люди (пациенты и персонал) были разбиты на двойки, в крайнем случае на тройки. Паате показалось, что все они о чем-то спорят, причем, когда двое говорили, третий обязательно молчал и слушал.

Внезапно Паата вспомнил шпионский роман, который он читал в детстве. Там у каждого агента СССР был западный контрагент, которые постоянно следили друг за другом и никогда не оставались одни. Иногда к ним присоединялся третий “контрагент” из Азии, но он или она постоянно молчали.

Паата быстро выпил воды, вернулся в палату, улегся на бок и заснул.

Через месяц его выписали. Есения не приехала его встречать. Паату встречает только его мать специально приехавшая из Тбилиси.

XIV

Динамо завоевало золото в тот год…

XV

После выписки, из-за морального и нервного истощения, Паата уже не мог в полной мере выполнять свою работу и решил уволиться в начале Апреля. Есения не поддержала это решение.

Брат Пааты, Эсарион, посоветовал ему обратить внимание на новую профессию программиста только начинающую появляться в СССР. И хотя профессия была новая для Союза, у Эсариона был богатый опыт в этой специальности, полученный им на оборонном предприятии Фархад (советское подразделение Министерства на Ближнем Востоке).

Через неделю после выписки, Паата поехал посетить своего друга Петра. Между ними состоялся следующий разговор:

• Ну как там твой отец, Петр? Снова в командировке?
• Да недавно уехал, у него новое назначение в Цюрих.
• Очень интересно и что он рассказывает про Швейцарию?
• Да много чего, например эта история про спящих агентов.
• Спящих агентов? Что это? Ленивые люди?
• Нет, по слухам это новая программа запада, когда человек не знает, что он агент, но подсознательно выполняет действия, направленные на подрыв нашего суверенитета.
• То есть как это подсознательно? Он что загипнотизирован?
• Это все что мне известно от отца.
• Да не парьcя! Если такие программы есть на западе, то и у наших наверняка тоже есть. Верно?
• Скверно. Суеверно.
• Характерно.

Друзья засмеялись.

XVI

Есения неожиданно без объяснения причин уходит от Пааты, и продлевает свою командировку в Китае.

Больше с Паатой они никогда не увидятся.

XVII

В 1978 году в Грузии происходят протесты против попытки ущемления прав грузинского языка. После чего Паата решает переехать в Тбилиси, движимый своей национальной гордостью.

Петр едет с ним.

Они устраиваются в ведущий вычислительный центр Грузинской ССР ‘Гурами’, ведь за плечами у них уже несколько лет работы программистами.

В тот же год, в Москве загадочно умирает агент А.

Настю усыновляет Маринка ( а вместе с ней возможно и советская власть).

XVIII

Паата проснулся как всегда поздно. Уже был одиннадцатый час…

Topology optimisation is a computational design method that finds the most efficient distribution of material within a defined design space, subject to specific loads and boundary conditions. It starts with a large, solid block of material (or void ) and an algorithm then iteratively removes (or adds) all the material that (does not) significantly contribute to the part’s stiffness and strength, forcing the structure to carry the load along the most direct paths. The result is an organic, lightweight, and high-performance geometry (often resembling bone or natural structures) that maximises the part’s structural efficiency, leading to significant weight and material savings, especially when combined with Additive Manufacturing (3D printing).

I have been doing research in applying reinforcement learning to topology optimisation since 2018. During this time, I have investigated Proximal Policy Optimisation; Hierarchical Reinforcement Learning; combination of genetic algorithms and RL; AlphaZero algorithm; and Pseudo 3D topology optimisation in order to make this approach practical for industry needs, [1–2].

In this article, I am going to explain one neat trick that I just came up with to improve compliance metrics and reduce run time of the algorithm. Namely, for 6 by 6 grid, I was able to reduce compliance from 19.92 to 18.16 (the lower the better), and reduce algorithm’s execution time from 19.337 min to 12.954 ( and even to 10.796) min (Fig.1 — Fig.2).

The idea of an method is as follows: for the first 300k (or 100k) iterations I do FEA to calculate compliance, but then I start comparing the current state (topology) with the best state (of best compliance) by cosine similarity and if it is less than threshold I do not do FEA and assign reward of zero, but if cosine similarity greater than threshold I do FEA and calculate the compliance in a usual way. The assumption is that better states are close to each other after some number of iterations.

Snapshot of the smart skipping trick can be seen below (for full implementation details please check my codebase).

# version 1
if  Tr.global_count <= 300_001:
    tmp, const = fast_stopt(self.args, self.x)
    self.reward = (1/tmp)**2
    if self.reward > Tr.max_reward:
        Tr.max_reward = self.reward
        Tr.best_state = self.x.reshape(6*6)
else:
    if cosine_similarity(self.x.reshape(6*6), Tr.best_state) < Tr.threshold:
        self.reward = 0
        const = 0
    else:
        tmp, const = fast_stopt(self.args, self.x)
        self.reward = 10*(1/tmp)**2

 # version 2
 if Tr.global_count <= 100_001:
     tmp, const = fast_stopt(self.args, self.x)
     self.reward = (1/tmp)**2
 else:
     if cosine_similarity(self.x.reshape(6*6), Tr.best_state) < (0.5+Tr.global_count/(2*ts))*Tr.threshold:
         self.reward = 0                                                            
         const = 0
     else:
         tmp, const = fast_stopt(self.args, self.x)
         self.reward = (1/tmp)**2
 
 if self.reward > Tr.max_reward:
     Tr.max_reward = self.reward
     Tr.best_state = self.x.reshape(6*6)

Also I experimented with sparse rewards which reduced run time down to ~7min for 6 by 6 grid.

# version 3 - sparse rewards
if done:
    tmp, const = fast_stopt(self.args, self.x)
    
    a1=1/tmp
    a2=1/const
    
    self.reward = 2*a1*a2/(a1+a2) 

It also allowed to get the following topologies for 8 by 8, 9 by 9, and 10 by 10 grids with pure DRL (Fig.3 — Fig.5):

That is more states than there are stars in the Universe! And all of this was done under 30 min wall-clock run time on my Mac M1 CPU.

Finally, I optimised 12x12 grid with sparse rewards, requesting ‘one island’,requesting ‘smooth’ geometry, and smart post-processing (removing elements with less than 2 neighbours), Fig. 6.

Preprint: https://www.researchgate.net/publication/398406554_Practical_topology_optimization_with_deep_reinforcement_learning

Codebase availability: https://github.com/gigatskhondia/gigala

To keep up with the project please visit https://gigala.io/

[1] https://www.researchgate.net/publication/389785241_Reinforcement_Learning_Guided_Engineering_Design_from_Topology_Optimization_to_Advanced_Modelling

[2] https://www.researchgate.net/publication/393164291_Pseudo_3D_topology_optimisation_with_reinforcement_learning

My recent years passion is applying reinforcement learning to engineering design of mechanical elements and components [1–2]. Starting from 2018, I am gradually but stubbornly researching this topic in the hope of unlocking the illimitable potential of artificial intelligence for the real world engineering applications.

On my bumpy road, I have investigated Proximal Policy Optimisation; Hierarchical Reinforcement Learning; combination of genetic algorithms and RL; AlphaZero algorithm; and even Pseudo 3D topology optimisation in order to make this approach practical for industry needs (Fig. 1 and Fig. 2).

But now I’m aiming for something bigger.

Believe it or not, I genuinely think that it is totally feasible to design a rocket engine end to end with my approach. What we need is physics simulator ( in the form of finite element models and rocket trajectory simulator, e.g. [3]), hierarchical reinforcement learning (potentially combined with genetic algorithms), and a personal computer (not sure that PC’s power is enough for this task, but at least it is how I am going to do this).

Using hierarchical reinforcement learning, where you have different abstraction levels or execution hierarchies of a design, with ‘managers’ (who do planning) and ‘workers’ who do the ‘actual work’, can lead to the systems of the full design cycle, where at the lowest level, the RL agent e.g. designs material. One abstraction level above, the RL agent designs a component (assembly of materials), one level above that, the RL agent produces a product (assembly of components).

At each level, the RL agent makes design decisions within its abstraction level, but all the levels communicate with each other during the training. At the end, the final design is produced in an end to end hierarchical fashion.

For the sake of example, let’s imagine we apply three abstraction levels to design our engine.

On level # 0, we would design individual components of the engine, like nozzle, for instance (Fig.3), by optimising their shape with reinforcement learning.

On level # 1, we would combine individual components into propulsion system, and optimise for the better placement of constituent elements, (Fig.4).

Finally, on level # 2, we would test our design within the framework of entire rocket, for vibration, for example ( Fig.5).

Importantly, all of these three abstraction levels would be designed simultaneously in parallel by a single RL agent in hierarchical fashion. The three levels would communicate with each other and produce the design end to end (e.g. as in [5]).

Early works of applying HRL to topology optimisation can be found in [1–2].

Some initial experiments on nozzle design can be seen on Fig. 6 and in my codebase (see below).

I do not know how to you, but to me this approach seems to be within reach and opens the door to space exploration. Now, would you press the button and get the result?

Updated Jan 29, 2026 with the link to most recent article: https://www.sciencepublishinggroup.com/article/10.11648/j.ajasr.20261201.11

Codebase availability: https://github.com/gigatskhondia/gigala

To keep up with the project please visit https://gigala.io/

[1] https://www.researchgate.net/publication/389785241_Reinforcement_Learning_Guided_Engineering_Design_from_Topology_Optimization_to_Advanced_Modelling

[2] https://www.researchgate.net/publication/393164291_Pseudo_3D_topology_optimisation_with_reinforcement_learning

[3] https://github.com/RocketPy-Team/RocketPy

[4] Rocket Propulsion Elements, George P. Sutton and Oscar Biblarz

[5] https://arxiv.org/pdf/1712.00948