Medium

Paul Deepakraj Retinraj

he

Pinned

Paul Deepakraj Retinraj

·

Jan 26, 2025

[InDepth] How DeepSeek R1 Model Works?

Architecture, Training, Benchmark Comparison, Limitations etc.

[InDepth] How DeepSeek R1 Model Works?

Pinned

In

AI Science

by

Paul Deepakraj Retinraj

·

Dec 30, 2022

Designing a Model Monitoring and Observability System

Model Monitoring/Observability Architecture, techniques, and solutions

Designing a Model Monitoring and Observability System

Pinned

Paul Deepakraj Retinraj

·

Jun 1, 2022

Designing Real-time Machine Learning Systems

Online machine learning model learning and inferences

Designing Real-time Machine Learning Systems

Paul Deepakraj Retinraj

·

Oct 24

The Agent Security Kernel, Part II: Identity, Information Flow, Containment and Evaluation

Bounding authority when the model is untrusted — Part II of II. Part I defined the threat model, the seven invariants (I0–I6) and the…

The Agent Security Kernel, Part II: Identity, Information Flow, Containment and Evaluation

Paul Deepakraj Retinraj

·

Oct 24

The Agent Security Kernel, Part I: Threat Model, Invariants and Architecture

Bounding authority when the model is untrusted — Part I of II. Part I covers the thesis, threat model, seven invariants, architecture and a…

The Agent Security Kernel, Part I: Threat Model, Invariants and Architecture

Paul Deepakraj Retinraj

·

Sep 27

Jev: Why AI Agents May Need Decision Models, Not Just Bigger LLMs

TypeSafe AI’s new System One model proposes a different role for AI in software: instead of generating text for people, it returns typed…

Jev: Why AI Agents May Need Decision Models, Not Just Bigger LLMs

Paul Deepakraj Retinraj

·

Aug 3

[In Depth] Part 3 — When Memory Gets Poisoned: Provenance, Trust, and Forgetting as Remediation

Every access check passed — and the lie still got in. Why enterprise agent memory needs an integrity layer, and how to build one.

[In Depth] Part 3 — When Memory Gets Poisoned: Provenance, Trust, and Forgetting as Remediation

Paul Deepakraj Retinraj

·

Aug 2

[In Depth] Part 2 — The Governed Context Graph: Access, Authority, and Trust for Enterprise Agent…

Why valid time isn’t enough — and how workload identity, relationship-based access control, and source authority decide who may read a…

[In Depth] Part 2 — The Governed Context Graph: Access, Authority, and Trust for Enterprise Agent…

Paul Deepakraj Retinraj

·

Aug 2

[In Depth] When “True” Has a Timestamp: Temporal Context Graphs for Enterprise Agent Memory

Why valid-time modeling — not better vector search — is what stops your agents from confidently acting on stale facts.

[In Depth] When “True” Has a Timestamp: Temporal Context Graphs for Enterprise Agent Memory

Paul Deepakraj Retinraj

·

Mar 22

[In Depth] Long-horizon agents and the trace flywheel: why data from agents will matter more than…

The next competitive moat in AI won’t be who has the best foundation model. It will be who has the richest corpus of agent execution traces…

[In Depth] Long-horizon agents and the trace flywheel: why data from agents will matter more than…

Paul Deepakraj Retinraj

Paul Deepakraj Retinrajhe

Distinguished Engg at Scale AI. Past: Salesforce and Oracle. AI Agents, Machine Learning & Artificial Intelligence. https://www.linkedin.com/in/pauldeepakraj/

Following

Help

Status

About

Careers

Press

Blog

Store

Privacy

Rules

Terms

Text to speech