Youness Ichen

Independent AI engineer designing and shipping agentic AI systems for SMEs and startups across finance, supply chain, HR and operations (Europe & US). 10+ projects delivered end-to-end, from scoping to production. Selected work: • Multi-agent financial analysis platform: LangGraph agents over market-data sources (FactSet / Capital IQ) with a multi-method valuation engine. • RAG chatbots with built-in lead qualification, embedded in client websites and pushed to the CRM (Claude API, PostgreSQL/pgvector, GoHighLevel). • AI prospecting agent: autonomous lead search, scoring and personalized outreach drafting (Agno, Exa, Gemini). • Read-only AI reporting assistants over business tools (CRM + team chat) delivering daily operational summaries. • Workflow automation and MCP integrations connecting LLMs to client systems. Stack: Python, LangGraph, CrewAI, Agno, Claude / OpenAI / Mistral APIs, MCP, RAG, FastAPI, Docker, PostgreSQL/pgvector, n8n.

Lead AI engineer on a production generative-AI assistant for a pension fund (regulated environment, sovereign hosting). • RAG chatbot in production: custom Python routing architecture, PostgreSQL/pgvector, bilingual French/Arabic. • PII protection: 76-rule bilingual sanitization pipeline (Presidio, spaCy, GLiNER, regex, fuzzy matching) + sensitive-document classification (~55 document types) with human validation after masking. • Security: guardrails tested against 269 Garak jailbreak attempts, 100% blocked; fail-closed behavior documented for the security audit. • On-premise LLMs: benchmarked 8 local models on CPU-only infrastructure; designed the local vs. cloud (Mistral / GPT-4o-mini) recommendation with semantic caching. • Evaluation & observability: RAG evaluation with Langfuse / Galileo / Arize Phoenix. • Deployment: private cloud VMs, CI/CD with GitHub Actions, nginx, SSL, Meta (WhatsApp) webhook integration.

Architecting and deploying an intelligent multi-agent system to 𝐚𝐮𝐭𝐨𝐦𝐚𝐭𝐞 𝐜𝐲𝐛𝐞𝐫𝐬𝐞𝐜𝐮𝐫𝐢𝐭𝐲 𝐫𝐞𝐜𝐨𝐧𝐧𝐚𝐢𝐬𝐬𝐚𝐧𝐜𝐞 𝐰𝐨𝐫𝐤𝐟𝐥𝐨𝐰𝐬 𝐟𝐨𝐫 𝐩𝐞𝐧𝐞𝐭𝐫𝐚𝐭𝐢𝐨𝐧 𝐭𝐞𝐬𝐭𝐢𝐧𝐠. Delivering: 𝐀𝐮𝐭𝐨𝐦𝐚𝐭𝐞𝐝 𝐑𝐞𝐜𝐨𝐧𝐧𝐚𝐢𝐬𝐬𝐚𝐧𝐜𝐞: Designed autonomous AI agents to independently execute complex reconnaissance tasks on web targets, eliminating manual bottleneck processes. 𝐃𝐲𝐧𝐚𝐦𝐢𝐜 𝐎𝐫𝐜𝐡𝐞𝐬𝐭𝐫𝐚𝐭𝐢𝐨𝐧: Built a custom architecture that seamlessly coordinates workflows, manages context, and handles data caching without human intervention. 𝐀𝐜𝐭𝐢𝐨𝐧𝐚𝐛𝐥𝐞 𝐈𝐧𝐭𝐞𝐥𝐥𝐢𝐠𝐞𝐧𝐜𝐞: Implemented automated, real-time reporting pipelines to deliver immediate insights directly to security teams. 𝐏𝐞𝐫𝐟𝐨𝐫𝐦𝐚𝐧𝐜𝐞 𝐎𝐩𝐭𝐢𝐦𝐢𝐳𝐚𝐭𝐢𝐨𝐧: Significantly reduced the execution time of vulnerability assessments while drastically improving the precision and reliability of the analysis.