<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Lucas Dahan on Medium]]></title>
        <description><![CDATA[Stories by Lucas Dahan on Medium]]></description>
        <link>https://medium.com/@lucas.dahan?source=rss-99061d9055e3------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*CMpk0rYEGMvbPgJ2oqhbgA.jpeg</url>
            <title>Stories by Lucas Dahan on Medium</title>
            <link>https://medium.com/@lucas.dahan?source=rss-99061d9055e3------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Mon, 16 Nov 2026 22:47:58 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@lucas.dahan/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Hands on LangGraph — Building a multi agent assistant]]></title>
            <link>https://medium.com/@lucas.dahan/hands-on-langgraph-building-a-multi-agent-assistant-06aa68ed942f?source=rss-99061d9055e3------2</link>
            <guid isPermaLink="false">https://medium.com/p/06aa68ed942f</guid>
            <category><![CDATA[langchain]]></category>
            <category><![CDATA[langgraph]]></category>
            <category><![CDATA[tutorial]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[genai]]></category>
            <dc:creator><![CDATA[Lucas Dahan]]></dc:creator>
            <pubDate>Wed, 26 Jun 2024 20:10:47 GMT</pubDate>
            <atom:updated>2024-07-15T10:43:16.321Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9qIXjWTv_tnApz8lW9g15A.jpeg" /></figure><h3>Hands on LangGraph — Building a multi agent assistant</h3><p>If you have been working on building a LLM product recently, you must have met and work with LangChain 🦜. It’s a great tool to build your first GenAI application and has democratized Retrieval Augmented Generation (RAG).</p><p>You can see LangChain as a friend who’s great at multitasking. But sometimes one friend is not enough to handle a huge project.</p><p>That’s where LangGraph appears. LangGraph enable the creation of interactions graphs between different “agent”. Simply, it works more like a master to organize group projects. The good part is that it facilitate a more flexible and scalable approach by making different GenAI agent working together. And the best part is that it’s really simple to create and add a new agent!</p><p>Now let’s dive in and get a practical guide to build your first LangGraph application.</p><h3>Starting a first LangGraph Project</h3><p>Before starting your project you have to take a step back and think of what do you want to achieve ?</p><p>For this tutorial, we want to build a simple application that is going to evaluate if a user prompt is technical or not. Regarding the result, the answer is going to be handle by a code specialist agent or a regular agent.</p><p>The figure below is a simple description of how it works.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PdB7J3LfrYosLjsB9lonNw.png" /></figure><p>While this example is really simple, it illustrates how easily the system can be expanded to accommodate more complex workflows. Each agent can leverage different models, prompt libraries, and maintain a degree of autonomy. That’s where LangGraph gets it strength: facility of creating and implementing new agents, allowing for rapid iteration and expansion of your GenAI Application.</p><p>Now let’s look at the different steps to build this LangGraph application.</p><p>First, setup your environment (<em>I suggest you to create a .venv</em>) with the proper packages:</p><pre>pip install langgraph langchain langchain-openai </pre><p>Now lets go to the first step, making our agents.</p><h3>Creating agents</h3><p>Let’s start by creating an agents.py file to define our three agents:</p><ul><li>The &quot;analysis&quot; agent for initial prompt classification</li><li>The &quot;code&quot; agent for code-related queries</li><li>The &quot;general&quot; agent for global topics</li></ul><pre>from langchain_openai import ChatOpenAI<br>from langchain.prompts import PromptTemplate<br>from langchain.chains import RetrievalQA<br>from langchain.llms import OpenAI<br><br># Creating the first analysis agent to check the prompt structure<br># This print part helps you to trace the graph decisions<br>def analyze_question(state):<br>    llm = ChatOpenAI()<br>    prompt = PromptTemplate.from_template(&quot;&quot;&quot;<br>    You are an agent that needs to define if a question is a technical code one or a general one.<br><br>    Question : {input}<br><br>    Analyse the question. Only answer with &quot;code&quot; if the question is about technical development. If not just answer &quot;general&quot;.<br><br>    Your answer (code/general) :<br>    &quot;&quot;&quot;)<br>    chain = prompt | llm<br>    response = chain.invoke({&quot;input&quot;: state[&quot;input&quot;]})<br>    decision = response.content.strip().lower()<br>    return {&quot;decision&quot;: decision, &quot;input&quot;: state[&quot;input&quot;]}<br><br># Creating the code agent that could be way more technical<br>def answer_code_question(state):<br>    llm = ChatOpenAI()<br>    prompt = PromptTemplate.from_template(<br>        &quot;You are a software engineer. Answer this question with step by steps details : {input}&quot;<br>    )<br>    chain = prompt | llm<br>    response = chain.invoke({&quot;input&quot;: state[&quot;input&quot;]})<br>    return {&quot;output&quot;: response}<br><br># Creating the generic agent<br>def answer_generic_question(state):<br>    llm = ChatOpenAI()<br>    prompt = PromptTemplate.from_template(<br>        &quot;Give a general and concise answer to the question: {input}&quot;<br>    )<br>    chain = prompt | llm<br>    response = chain.invoke({&quot;input&quot;: state[&quot;input&quot;]})<br>    return {&quot;output&quot;: response}</pre><p>In this example, all three agents use OpenAI’s model. However, LangGraph’s allows for the integration of various models, parameters, and tasks within each agent. You can setup these agents with more complex prompts and even implement Retrieval-Augmented Generation (RAG) techniques to improve their capabilities.</p><p>Now that we have the agents, lets create the Graph.</p><h3>Building the Graph</h3><p>With our agents defined, we’ll create a graph.py file to orchestrate their interactions. The basic graph structure in LangGraph is really simple, here we are going to use three steps:</p><ul><li>Associate “nodes” to our agents to find them in the Graph</li><li>Define “activation conditions” of the agents</li><li>Establish a “workflow” for our graph with the entry point and stops</li></ul><pre><br>from langgraph.graph import StateGraph, END<br>from typing import Annotated, TypedDict<br>from agents import analyze_question, answer_code_question, answer_generic_question<br><br>#You can precise the format here which could be helpfull for multimodal graphs<br>class AgentState(TypedDict):<br>    input: str<br>    output: str<br>    decision: str<br><br>#Here is a simple 3 steps graph that is going to be working in the bellow &quot;decision&quot; condition<br>def create_graph():<br>    workflow = StateGraph(AgentState)<br><br>    workflow.add_node(&quot;analyze&quot;, analyze_question)<br>    workflow.add_node(&quot;code_agent&quot;, answer_code_question)<br>    workflow.add_node(&quot;generic_agent&quot;, answer_generic_question)<br><br>    workflow.add_conditional_edges(<br>        &quot;analyze&quot;,<br>        lambda x: x[&quot;decision&quot;],<br>        {<br>            &quot;code&quot;: &quot;code_agent&quot;,<br>            &quot;general&quot;: &quot;generic_agent&quot;<br>        }<br>    )<br><br>    workflow.set_entry_point(&quot;analyze&quot;)<br>    workflow.add_edge(&quot;code_agent&quot;, END)<br>    workflow.add_edge(&quot;generic_agent&quot;, END)<br><br>    return workflow.compile()</pre><p>As you can see, the structure is simple and easily adaptative. You can add new agent, change the conditions, modify the workflow and create way more complex structure to handle specifics GenAI uses cases.</p><p>Finally we are going to try this first application!</p><h3>Launching the Program</h3><p>Finally, we’ll create a main.py file to bring our LangGraph project to life. This file initializes the graph and provides an interface for our first interaction!</p><pre>import os<br><br>#Embedding API keys directly in your code is definilty not secure or recommended for production environments.<br>#Always use proper key management practices.<br>os.environ[&quot;OPENAI_API_KEY&quot;] = &quot;your-openai-key&quot;<br><br>from graph import create_graph<br>from typing import Annotated, TypedDict<br>from langgraph.graph import StateGraph, END<br><br>class UserInput(TypedDict):<br>    input: str<br>    continue_conversation: bool<br><br>def get_user_input(state: UserInput) -&gt; UserInput:<br>    user_input = input(&quot;\nEnter your question (ou &#39;q&#39; to quit) : &quot;)<br>    return {<br>        &quot;input&quot;: user_input,<br>        &quot;continue_conversation&quot;: user_input.lower() != &#39;q&#39;<br>    }<br><br>def process_question(state: UserInput):<br>    graph = create_graph()<br>    result = graph.invoke({&quot;input&quot;: state[&quot;input&quot;]})<br>    print(&quot;\n--- Final answer ---&quot;)<br>    print(result[&quot;output&quot;])<br>    return state<br><br>def create_conversation_graph():<br>    workflow = StateGraph(UserInput)<br><br>    workflow.add_node(&quot;get_input&quot;, get_user_input)<br>    workflow.add_node(&quot;process_question&quot;, process_question)<br><br>    workflow.set_entry_point(&quot;get_input&quot;)<br><br>    workflow.add_conditional_edges(<br>        &quot;get_input&quot;,<br>        lambda x: &quot;continue&quot; if x[&quot;continue_conversation&quot;] else &quot;end&quot;,<br>        {<br>            &quot;continue&quot;: &quot;process_question&quot;,<br>            &quot;end&quot;: END<br>        }<br>    )<br><br>    workflow.add_edge(&quot;process_question&quot;, &quot;get_input&quot;)<br><br>    return workflow.compile()<br><br>def main():<br>    conversation_graph = create_conversation_graph()<br>    conversation_graph.invoke({&quot;input&quot;: &quot;&quot;, &quot;continue_conversation&quot;: True})<br><br>if __name__ == &quot;__main__&quot;:<br>    main()</pre><p>And… That’s it! With a simple python main.py command, your first LangGraph project is up and running.</p><p>This version explicitly describes the choices made by the graph. The idea is to provide insights to help you refine your agents prompt and workflow.</p><h3>To wrap-up!</h3><p>First, thank you for reading this article! I really wanted to try a first “technical” post specially on LangGraph.</p><p>From my perspective, LangGraph liberates developers from the constraints of LangChain’s function cascades. The modular approach to orchestrating multiple agents is great to handle more complexe problems and better output quality. Plus, it’s honestly super quick to add a new agent in the graph!</p><p>It opens the path to way more complex (multimodal or not) architecture. With possibilities to ensure more consistent, relevant and safe application to your GenAI product!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=06aa68ed942f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to achieve ROI from your AI Projects]]></title>
            <link>https://medium.com/@lucas.dahan/how-to-achieve-roi-from-your-ai-projects-5ab67de31a5f?source=rss-99061d9055e3------2</link>
            <guid isPermaLink="false">https://medium.com/p/5ab67de31a5f</guid>
            <category><![CDATA[data]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[product-manager]]></category>
            <category><![CDATA[product]]></category>
            <dc:creator><![CDATA[Lucas Dahan]]></dc:creator>
            <pubDate>Wed, 29 May 2024 12:09:14 GMT</pubDate>
            <atom:updated>2024-05-30T08:20:37.643Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5HQD3yGHhH_P_jwzYadR9w.jpeg" /></figure><p><em>Recognizing the need to treat AI as a component of a product meant to deliver value is the first step towards a positive ROI.</em></p><p>Artificial intelligence is at the heart of companies’ strategic concerns. Despite this ambition, a strong observation emerges: according to BCG, 90% of companies do not go beyond simple experiments concerning artificial intelligence. This statistic raises a question: how to transform costly experiments into products generating significant ROI?</p><p>The shift from a project approach to a product approach is the decisive factor for successfully making this transition.</p><p>This series of articles offers you a practical guide to achieving this.</p><h4>The ROI Issue in AI Projects</h4><p>The adoption of AI is seen as a lever for growth and innovation. Acquiring the technology and implementing it becomes a priority. However, transitioning from the experimentation phase to the production phase proves to be a major challenge for most companies. The experimentation phases, although necessary for technology integration, are often costly and do not bring direct ROI.</p><p>The reason? An approach primarily focused on technology rather than creating value for the company and its customers.</p><p>Companies that have successfully made this transition have one thing in common: the adoption of a product approach from the beginning of their AI projects. This methodology aims to transform experimental use cases into tangible, measurable assets that generate value. By organizing execution with product teams, initiatives can move from idea to profitable innovation. Three key steps will allow you to make this transition sustainably.</p><h4>The Discovery Phase to Validate Your AI Projects</h4><p>The Discovery phase allows you to focus on the needs of your users. This first step ensures the relevance of your projects and allows you to build a robust business plan to turn your initiative into an actionable roadmap.</p><p>Our article “<a href="https://medium.com/@lucas.dahan/do-you-need-ai-introduce-discovery-into-your-projects-ebd3f6b9f0c1">Do You Need AI? Introduce Discovery in Your Projects</a>” details all the elements to succeed in this first phase.</p><h4>The Build Phase Turns Projects into Integrated, Sustainable Products</h4><p>The Build phase allows you to transform your AI proof of concept into a robust, value-generating product. By controlling your costs (FinOps) and the robustness of your models (MLOps), you can set up the entire system necessary for developing your AI product.</p><p>The article “<a href="https://medium.com/@lucas.dahan/turn-your-ai-pocs-into-sustainable-products-with-build-5a46072a6a1c">Transform Your AI Proof of Concepts into Sustainable Products with the Build Phase</a>” provides all the necessary details to understand and start this construction phase.</p><h4>The Run Phase Ensures the Evolution and Success of Your New AI Products</h4><p>Integrating the Run phase in an AI product framework is not limited to simple maintenance. This step is crucial to avoid model drifts and to leverage the full potential of the feedback loop from your AI products. It establishes a virtuous cycle of continuous improvement to evolve your product according to new user expectations.</p><p>The fourth article “<a href="https://medium.com/@lucas.dahan/make-your-ai-product-evolve-to-ensure-its-success-thanks-to-the-run-5b780240bbc6">Make your AI product evolve to ensure its success thanks to the Run</a>” will guide you on the elements to implement to ensure the successful evolution of your AI products.</p><p>By following these three phases, you will ensure that you prioritize the right topics that will generate sustainable returns on investment.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5ab67de31a5f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Make your AI product evolve to ensure its success thanks to the Run.]]></title>
            <link>https://medium.com/@lucas.dahan/make-your-ai-product-evolve-to-ensure-its-success-thanks-to-the-run-5b780240bbc6?source=rss-99061d9055e3------2</link>
            <guid isPermaLink="false">https://medium.com/p/5b780240bbc6</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[product]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[data]]></category>
            <dc:creator><![CDATA[Lucas Dahan]]></dc:creator>
            <pubDate>Wed, 29 May 2024 12:05:31 GMT</pubDate>
            <atom:updated>2024-05-30T08:21:13.762Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zVdOnpvzIzD_y7G3GRrNJQ.jpeg" /></figure><p><em>The run phase isn’t just about maintaining your AI products, it’s about making them evolve to ensure a virtuous circle.</em></p><p>This article is the third and final in a series dedicated to the application of product management to artificial intelligence. Having explored the crucial “Discovery” and “Build” phases, we will now focus on the “Run” phase and its importance for the success and longevity of your AI products.</p><p>The “Run” phase ensures that AI models remain relevant in the face of evolving data and user needs. It optimizes the feedback loop for continuous product improvement, creating a virtuous circle of innovation and value creation.</p><h4>Drift: when robustness becomes the enemy of your models</h4><p>AI products, however sophisticated, are not immune to drift. Indeed, they rely on data that, by its very nature, evolves over time. Customer habits change, the market transforms, and the world around us is in constant flux. Without proper MLOps management during the Run phase, these evolutions can cause models to drift, rendering them obsolete and ineffective.</p><p>The Run phase, which is designed not to let your product “live on its own”, counteracts this drift by constantly monitoring model performance and adjusting parameters as necessary. Monitoring and alert tools enable you to detect weak signals of drift and intervene quickly, or even proactively, to maintain the model’s relevance.</p><p>The Covid-19 example illustrates this phenomenon perfectly. The pandemic disrupted consumer habits, rendering obsolete many AI models that had not anticipated these changes. Companies that were able to adapt through effective “run” management were able to maintain the performance of their products and even seize new opportunities.</p><h4>Feedback loop: get the most out of your AI products</h4><p>One of the greatest strengths of AI products is their ability to learn and improve through a feedback loop. This loop enables new data and user feedback to be integrated to refine the model and make it perform better.</p><p>The Run phase plays an essential role in optimizing this feedback loop. It’s important to remember that an AI product is not simply a model, but a complete system that includes data management, infrastructure and user interfaces. If data quality is essential at product launch, it is just as important, if not more so, during the Run phase.</p><p>During this phase, it is essential to prioritize the development of features that improve data access and quality. This can include monitoring tools, automated test systems, the opening up of promising new streams, or even features that enable users to provide feedback on the relevance of results.</p><p>By maximizing the feedback loop, you enable your AI product to learn and adapt continuously, which is essential for maintaining its performance and added value over the long term.</p><h4>Virtuous circle: a successful product is an evolving product</h4><p>The success of an AI product depends on its ability to evolve and adapt to user needs. The first step in discovery is to focus on users. This run phase sets in motion a virtuous circle of continuous improvement by re-launching the Discovery phase, then the Build phase…</p><p>Including time devoted to analyzing user data and behavior in the run phase helps identify new opportunities to feed the Discovery phase and explore new use cases. The “Build” phase is then used to develop and integrate these new functionalities, before deploying them in production and monitoring them in the “Run” phase. The aim is to bring your product to life.</p><p>One of the most emblematic examples is Facebook. Over the years, the social network has evolved from a simple networking platform to a complete ecosystem of products and services. This ability to adapt and innovate has been the key to its success over the years.</p><h4>The Run: the key stage in the evolution of your AI products</h4><p>The “Run” phase is much more than simple maintenance: it’s the driving force behind the evolution and longevity of your AI products. By combating model drift, optimizing the feedback loop and creating a virtuous circle of continuous improvement, you ensure the long-term relevance and performance of your products.</p><p>While the synergy between the “Discovery”, “Build” and “Run” phases enables the implementation of an efficient organization and a solid technological environment, the complete success of these projects cannot be achieved without a profound cultural transformation. Training employees, supporting them through change and creating a culture of continuous learning around AI are essential elements in guaranteeing the adoption and sustainability of your future products!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5b780240bbc6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Do you need AI? Introduce discovery into your projects]]></title>
            <link>https://medium.com/@lucas.dahan/do-you-need-ai-introduce-discovery-into-your-projects-ebd3f6b9f0c1?source=rss-99061d9055e3------2</link>
            <guid isPermaLink="false">https://medium.com/p/ebd3f6b9f0c1</guid>
            <category><![CDATA[innovation-management]]></category>
            <category><![CDATA[product]]></category>
            <category><![CDATA[data]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Lucas Dahan]]></dc:creator>
            <pubDate>Wed, 29 May 2024 12:03:57 GMT</pubDate>
            <atom:updated>2024-05-30T08:21:53.733Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7wF0SSGcei6HsFNrZE6ofA.jpeg" /></figure><p><em>The first step in securing a return on investment is to ensure that the use of AI is appropriate, thanks to a discovery phase.</em></p><p>Not all your needs will be solved with Artificial Intelligence: relevance, cost, processing urgency or return on investment are among the indicators that need to be assessed before you take the plunge. To evaluate these elements and build a solution, it is necessary to move from a project approach to a product approach, using the preliminary study and planning phase: Discovery.</p><p>Integrating the Discovery phase into your AI projects is strategically crucial. It ensures that the innovations developed genuinely meet specific needs, while optimizing the chances of commercial success. It is the first essential tool for identifying high-potential projects and securing investments for the Build phase.</p><h4>Understanding user needs: ensuring your project’s relevance</h4><p>The Discovery process begins with an in-depth analysis of user needs and expectations.</p><p>This crucial step ensures that AI POCs, which are often experimental and innovative, are truly aligned with concrete needs and offer significant added value. This avoids developing solutions in search of problems, by ensuring that each AI project proposes a relevant response to real business challenges.</p><h4>The value hypothesis: build a convincing business plan</h4><p>Defining value hypotheses is at the heart of Discovery. This phase aims to clarify how the AI product meets an identified need and will contribute to the creation of value for your company.</p><p>Building a solid business plan, articulating the expected benefits and spin-offs, is essential to justify the necessary investments and determine the KPIs, the key success indicators. This provides a reliable basis for project management and evaluation.</p><h4>An agile roadmap: guide innovation with precision</h4><p>The creation of a strategic roadmap, the final element of Discovery, is crucial to the agile management of your future AI products.</p><p>This plan details the key stages of development, enabling you to maintain a consistent direction while adapting to technological evolutions and market opportunities. It ensures a structured, targeted progression from POCs to finished products, in harmony with your long-term objectives.</p><p>You can also read my article on the “Build” phase, which enables you to create a robust product and validate investments: <a href="https://medium.com/@lucas.dahan/turn-your-ai-pocs-into-sustainable-products-with-build-5a46072a6a1c">Turn your AI PoCs into sustainable products with Build</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ebd3f6b9f0c1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Turn your AI PoCs into sustainable products with Build]]></title>
            <link>https://medium.com/@lucas.dahan/turn-your-ai-pocs-into-sustainable-products-with-build-5a46072a6a1c?source=rss-99061d9055e3------2</link>
            <guid isPermaLink="false">https://medium.com/p/5a46072a6a1c</guid>
            <category><![CDATA[product-manager]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[product]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[data]]></category>
            <dc:creator><![CDATA[Lucas Dahan]]></dc:creator>
            <pubDate>Wed, 29 May 2024 11:58:12 GMT</pubDate>
            <atom:updated>2024-05-30T08:22:17.131Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*eMaLrOp63YxsgiGO8X-20g.jpeg" /></figure><p><em>The second Build stage enables you to build a robust product and validate your investments.</em></p><p>To successfully turn AI into a strategic opportunity, it’s essential to change perspective and view AI as a product, not a project.</p><p>This article, one of a series on product management applied to AI, explores the “Build” phase, which transforms prototypes into robust, scalable and value-generating AI systems. It enables you to move beyond the POC phase by providing the environment needed to build an AI system, while controlling costs (thanks to FinOps) and the robustness of your models (thanks to MLOps).</p><h4>Beyond the prototyping phase: the challenge of production</h4><p>The success of your Discovery phase is not always enough to move beyond the prototyping phase. POCs are often developed without a clear vision of how they will be put into production and integrated into the information system. What’s more, the technical challenges involved in putting an AI system into production, such as data management, scalability and security, are often underestimated. Finally, the lack of established governance for AI projects can lead to delays, budget overruns and quality issues.</p><p>The “Build” phase enables you to overcome these obstacles by taking a structured approach and focusing on building a sustainable, scalable AI system. The approach to an AI product as a system is essential, and must integrate release, integration with the information system, and lifecycle management of the entire system.</p><h4>Build, or the art of mastery: control and stability</h4><p>Build is based on the principle of control. It enables you to control the various aspects of your AI products and guarantee their long-term stability. Two key elements contribute to this control: FinOps and MLOps.</p><p>FinOps provides a framework for optimizing cloud spending and ensuring the profitability of AI products. It enables you to track and analyze costs, identify optimization opportunities and make informed resource allocation decisions. In this way, you can ensure that your AI products remain economically viable and do not exceed the value assumptions defined during Discovery.</p><p>MLOps aims to industrialize the lifecycle of AI models. It enables you to manage model volatility by automating monitoring, drift detection and updating processes. A good example is model versioning. It enables you to monitor the evolution of your models and revert to a previous version in the event of a problem. Thanks to the many practices of MLOps, you can guarantee the performance and reliability of your AI systems in production.</p><h4>The systems approach as a key to success: integration and adaptability</h4><p>The systems approach is a key element of the Build phase. It focuses not just on your models, but on the systems that guarantee their success. To do this, it takes into account all the elements necessary for the success of an AI system: infrastructure, data, models, processes and user interfaces. This approach ensures system consistency and performance. It ensures that all AI system components work harmoniously together, and that the system is capable of meeting user needs.</p><p>Continuous integration (CI) and continuous deployment (CD) are essential practices for building an agile and adaptable AI system. CI/CD enables you to automate the testing, deployment and update processes, providing your teams with a breeding ground for collaborative development while ensuring the quality of functionality.</p><p>Thanks to the systemic approach and CI/CD practices provided by Build, you can ensure that your AI systems remain relevant and efficient in an environment designed for evolution.</p><h4>Build: the best way to perpetuate your AI products</h4><p>By considering AI as a system, the “Build” phase helps to control financial costs and model drift, thanks to FinOps and MLOps. This phase provides a robust, scalable environment for development teams, favoring continuous integration and ongoing deployment. This is a crucial step in turning the promise of AI into tangible reality.</p><p>You can also read my article on the “Run” phase, which enables you to stabilize AI products and create an evolutionary virtuous circle, ensuring the durability of your innovations: <a href="https://medium.com/@lucas.dahan/make-your-ai-product-evolve-to-ensure-its-success-thanks-to-the-run-5b780240bbc6">Make your AI product evolve to ensure its success thanks to the Run</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5a46072a6a1c" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>