Freelancer profile translated to English.
Back to original languageAbout Martin
In 2026,a language model can become the equivalent of an autonomous collaborator**: knowing your files, your procedures, your clients, and responding with the precision of someone who has worked for you for years. The difference is no longer in the model's power or how you talk to it, but in **the context architecture made available to it.This is my job: context engineering. The same model will be exceptional or useless depending on the context it's fed.
- An example.For an outreach agency, I retrieved hundreds of scattered sources: call transcripts, produced documents, sales files. Valuable material, but which the disorder reduced to files piled up on a cloud. I transformed it into a locally hosted, indexed, and structured knowledge base, queryable with the model of choice. Itresponds accurately on all the company's activities, citing its sources, because the context was designed for it.Not a generic model with documents attached: an architecture where the right information surfaces at the right time.
This is what I build for my clients: systems that respond from your data (PDFs, Drive, Notion, tickets) by citing their sources, models integrated into your product, agents that handle a business task end-to-end. The common thread: responding accurately, recognizing when the agent doesn't know instead of making things up, and performing under real-world conditions.
Today I work for a French AI firm that delivers agents in production in MedTech. In parallel, I designed and put two SaaS products into production in a few months, where a traditional team would have taken years. The real issue isn't that they exist, but what made it possible: **context engineering that makes AI work like an entire team**, without compromising on rigor or structure.
French
Native or bilingual
English
Native or bilingual
Remote only
Primarily works remotely
Experience
- Cabinet d'IA Francais (Medtech)AI EngineerTECHOctober 2025 - Today (8 months)France
Design and optimization of AI agents delivered in production for clients in the medical sector.
I am involved in the development of agents and tasks, prompt engineering, monitoring, and observability (Langfuse). Working in TypeScript within a large-scale orchestrated agent architecture.Concrete deliverables:- End-to-end outreach agents: sourcing from multiple sources (academic research, company data, buying window signals, contact details), writing personalized messages, and managing sending campaigns to thousands of leads.
- Competitive intelligence agent operating autonomously on WhatsApp, OCR, adding leads to CRM, information retrieval...
- Sales call analysis agent: each call evaluated according to best social engineering practices, with an actionable summary produced at the end of the call for continuous improvement of the sales team's performance.
- Carots.frAI developerApril 2026 - Today (2 months)
CTO/Technical Lead at Carots — Pre-revenue SaaS
I lead the entire technical aspect of our SaaS, starting from scratch and bringing it to production in two months: a codebase of over 150,000 lines (Dart and SQL), currently used daily by three stables, with approximately 90 daily active users. The product is intentionally in private beta while we refine it before public launch.- The challenge isn't just writing code, but making this codebase operable by a team. I implemented a complete production workflow on Linear connected to Claude Code and a suite of skills I developed, structuring the process from idea to code. I created an agent, "Jean," which translates requests from my two non-technical co-founders into well-defined, ready-to-implement tickets: they fully contribute to the product without touching a line of code. I even trained them to manage their own projects on Linear using this system.
This is my real deliverable: an organizational discipline that allows one person to maintain a complex, prod-ready codebase, readable and manageable by non-technical profiles, and evolve it at the pace of a team. Exactly what I can implement for a client: an organization where AI and existing personnel produce software together, cleanly and sustainably. - Prestance.ioRAG engineerMarch 2026 - Today (3 months)
Design and solo production of an AI CV analysis SaaS, from architecture to interface, with its own acquisition channel
- The analysis pipeline combines several models based on their roles. An initial pass with Claude Haiku identifies missing aspects of a CV and selects the relevant reference article; user responses are then passed to Claude Sonnet for in-depth analysis. CVs are read by extracting information from valid PDFs, with an OCR fallback for non-standard formats, ensuring no document is rejected.
- RAG relies on vector search (pgvector, embeddings) that injects the most relevant editorial content into the model: each analysis is thus fed by a knowledge base that I maintain myself.
- This base comes from the acquisition channel I built: daily SEO publishing, driven by a workflow I optimize weekly using Google Search Console data and user behavior via the analytics tool. The content that attracts visitors is the same that refines the quality of the analyses. Closed loop.
Reliability treated as a priority: safeguards against fabricated responses, input anti-injection security, cost control (caching, injection limits).
Recommendations
Be the first to recommend Martin
Help this freelancer shine by sharing your experience working together.
These freelancer profiles also match your criteria
Agatha Frydrych
Backend Java Software Engineer
4.7
(3)
2
Baptiste Duhen
Fullstack developer
4.6
(4)
5
Amed Hamou
Senior Lead Developer
4
(2)
7
Audrey Champion
Web developer
4.3
(3)
4