AI Engineer - LLM agents in production, MCP (Model Context Protocol), RAG, MLOps, Python

2/2026 – 9/2026

Open source contributions around the MCP (Model Context Protocol) ecosystem. Merged contribution to the official MCP conformance suite, a Linux Foundation project (github.com/modelcontextprotocol/conformance, PR #492): check for deterministic tools/list ordering on the 2026-07-28 revision, twelve tests added, reviewed and merged by a project maintainer. Published npm packages, designed and released single-handed: mcp-migrate-delta (Apache-2.0), replays the official conformance suite before and after the move from MCP 2025-11-25 to 2026-07-28, 36 failing scenarios down to 0; a2a-to-mcp, active entry in the official MCP registry, one A2A (Agent2Agent) v1.0.1 agent becomes four MCP tools, 241 passing tests; mcp-opentelemetry (MIT), W3C Trace Context propagation, 50 conformance scenarios identical with and without instrumentation. HiveMind, shared memory for coding agents over MCP (github.com/AmirK-S/HiveMind): seven-tool MCP server, FastAPI, PostgreSQL with pgvector, Redis, Celery, LangChain and CrewAI wrappers, upgraded to protocol revision 2026-07-28, 69 tests in CI. Double, source-grounded knowledge base: MCP server and evaluation set, Python, FastAPI.

Postgresql, Continuous Integration, Langchain, Git, Python, Typescript

2/2026 – offen

Own product, designed, built and released single-handed: TTP (Talk To Paste), a desktop dictation app. Native macOS and Windows app (Tauri 2, Rust and React), version 3.1.6 shipped, paid licences: an operated product, not a demo. Transcription in under two seconds, LLM clean-up pass, personal dictionary. Apple-signed and notarised builds, signed Windows installers, minisign-verified updates, API keys stored in the OS keychain, published threat model.

Continuous Integration, Typescript, React (JavaScript library)

12/2025 – offen

Own product, designed and developed single-handed: Next.js platform (VideoForge / NicheRadar) for niche research and video production assistance. 12-module technical masterclass (140+ pages) produced and published.

Full-Stack

10/2025 – 9/2026

Designed a multi-agent backend for a customer-facing assistant, then designed and evaluated a contract-answering agent on an internal orchestration platform.

Deterministic guardrails rather than decisions delegated to the model: router to five specialist agents with a deterministic fallback, server-side claim-form state machine, a bounded tool loop over sixteen strict-schema functions, SSE streaming. Python, FastAPI, six dependencies, the whole control flow written by hand.

Built an evaluation harness that makes two campaigns comparable: versioned run registry, prompt fingerprint written into every generated artefact, single-variable A/B runs, a 353-question business-labelled test set.

Built a readability scoring engine for contract PDFs: layout extraction, linguistic analysis, forty detectors in separate files, thresholds externalised as JSON.

API-Entwickler, Python, Softwareentwickler

6/2025 – 10/2025

AI trainer and speaker, automation audits, for Arkel (training company, France), under DENEM Labs (my independent activity). Delivered in-company generative AI training sessions for business teams, with sessions delivered at SNCF, Tefal, Lacoste and Image7. Ran automation audits: process mapping, identification of high-impact automation candidates, 30 to 60 to 90 day roadmaps, up to the implementation of automation workflows.

Generative KI, Large Language Models, Digitalisierung

9/2024 – offen

RAG system for financial reports with structured extraction and summary generation, built under DENEM Labs (my independent activity). LangChain, Azure.

Document Retrieval, Langchain, Microsoft Azure

9/2024 – offen

MCP (Model Context Protocol) server for a commodities trading platform (ErisX), built under DENEM Labs (my independent activity). Python.

Python

9/2024 – offen

Intelligent PDF processing pipeline built under DENEM Labs (my independent activity): OCR, LLM extraction, and integration with Slack and Google Drive. LangGraph, Azure.

Python, Microsoft Azure

9/2024 – offen

Freelance AI engineering under DENEM Labs (my independent activity): design and delivery of generative AI systems for large accounts, mid-caps and SMEs. Multi-tool conversational agent with persistent memory (PostgreSQL) and Qdrant retriever; MCP (Model Context Protocol) server for a commodities trading platform (Python); document indexing pipelines (extraction, semantic chunking, embeddings, batch upserts to Pinecone and Qdrant), neural reranking (Cohere) on Supabase; RAG over financial reports with structured extraction and summaries (LangChain). A 40 plus stage knowledge processing chain (multi source ingestion, LLM enrichment, conditional routing, error handling and resume). Clients and principals named as credentials: Arkel (in company AI training, delivered at SNCF, Tefal, Lacoste and Image7) and H and Associes (design and production of the Claude for architects training programme). Detailed content of each engagement is under confidentiality.

Postgresql, Document Retrieval, Generative KI, Langchain, Large Language Models, API-Entwickler, Docker, Python

10/2023 – 9/2024

Data & AI Consultant, ENGIE (energy group, France). Built a production RAG over an R&D corpus of 1,000+ documents (LangChain, Azure OpenAI, Azure AI Search, vector indexing). Built a continuous ingestion system to Azure Blob Storage with preprocessing and validation.

Data Science, Langchain, Python

12/2021 – 10/2022

Rust Developer, SUN ZU Lab (market making and Web3 tooling studio, Paris, fintech). Designed domain models for order-book events and wrote the technical documentation. Wrote Python monitoring and alerting scripts for failure detection, ensuring 99.9% uptime.

Git, Python, Softwareentwickler