Ahmed Hajjem

Ahmed Hajjem

Forward Deployed AI Engineer
Tunisia
English, Arabic, French
Full-time Hybrid

About Me

Forward Deployed AI Engineer with experience delivering AI products from requirements to running systems, including deployment, hosting, observability, and incident response. Has built retrieval systems, legal research p…

Experience

Forward Deployed AI Engineer

Kairos Impulse · Carthage , Tunis , Tunisia
Jun 2026 - Present · 4 months

<p>Embedded with 3+ client organisations as their technical counterpart.</p><p> Ran discovery sessions, translated business problems into technical requirements, and delivered against them, typically 2 to 3 weeks from first session to a working system, then iterating with the client.</p><p> Shipped a bespoke Flutter app for La Brosse, Tunisia’s largest brush manufacturer, that reads and translates lettres de change on-device with a fine-tuned vision-language model, so sales staff key the data into their ERP in a fraction of the time.</p><p> Built an operations management system for Laurus Labs, an Indian pharmaceutical and biotechnology group, covering their operators and day-to-day operations.</p><p> Took AI products from requirement to running system: build, deploy, host and operate them on client infrastructure, covering rollout, security hardening, observability and incident response.</p><p> Owned models and LLMs in production end to end: deployment, monitoring, performance tuning and industrialisation of inference.</p>

AI Consultant & Developer

Legal Paper · Tunis, Tunisia
Aug 2025 - May 2026 · 8 months

<p>Built the internal legal research platform for Ferchiou & Associés, the largest law firm in Tunisia, giving lawyers unified search across 40,000+ laws, contracts and case documents at 84% search accuracy and relevance.</p><p> Cut legal research time by 98% and operational cost by 42%.</p><p> Combined Elasticsearch with Graph RAG on Neo4j for hybrid retrieval, tracking the version history and time validity of every document.</p><p> Engineered the document ingestion pipeline end to end, with OCR models fine-tuned per document type.</p><p> Shipped a legal assistant chatbot over both retrievers, running on open-source LLMs fine-tuned to the firm’s domain with QLoRA.</p><p> Built a Microsoft Outlook add-in routing mail automatically across approximately 1,900 legal mailing groups in both French and Arabic, classifying 5,000+ emails in its first months in production and cutting classification time by 76%.</p><p> Paired Azure AI Search fuzzy matching with semantic search on pgvector, using embeddings fine-tuned on legal language.</p><p> Designed and shipped a timesheet-management agent in Copilot Studio, aggregating lawyer activity from 5 Microsoft 365 sources through the Microsoft Graph API.</p><p> Led two engineers on the Knowledge Hub, owning architecture, code quality and delivery, while running 4+ concurrent production projects with weekly demos to partner-level stakeholders at the firm.</p>

AI Engineer (contract)

Jade Advisory · Tunis, Tunisia
Jan 2024 - Sep 2024 · 8 months

<p>Built a retrieval system over the firm’s archive of PPP feasibility studies: multi-representation indexing with separate text and table retrievers fused by Reciprocal Rank Fusion, plus re-ranking.</p><p> Added table-aware extraction with LLM-generated summaries, keeping financial tables legible to the model.</p><p> Benchmarked candidate models and landed on a hybrid of Gemini 1.5 Pro and locally hosted Mistral 7B, giving 15% better results than any single model or alternative combination.</p><p> Handled French-language sources with a local translation step.</p><p> Evaluated the pipeline with RAGAS and instrumented it with Langfuse for tracing and cost tracking.</p><p> Delivered a second system turning Excel financial tables into LLM-written report sections exported to Word, with a human review stage.</p><p> It now generates approximately 40% of Jade’s PPP feasibility reports.</p><p> Worked as part of a two-engineer team.</p>

PROJECTS

Troviale: API gateway for artificial intelligence

Own Project · https://troviale.com/
Duration : 01-Feb-2026 - 01-Sep-2026

<p>◦ Architecture: Built on Next.js and LiteLLM, then architected around them into a production-ready multi-tenant platform: API-key authentication, per-key usage metering and a 5-layer security model. ◦ Scale: Serves 1,000+ API calls per day in production. ◦ Infrastructure: Deployed on AWS EC2 with Docker, Traek, PostgreSQL, Redis and automated backups. Designed, shipped and operated single-handed. </p>

Skills

Python TypeScript JavaScript SQL LangChain LangGraph Next.js React Node.js Streamlit RAG AI agents prompt engineering open-source LLMs fine-tuning QLoRA PEFT LiteLLM RAGAS Langfuse hybrid search Graph RAG Neo4j Elasticsearch rank fusion re-ranking multi-representation indexing embeddings PostgreSQL pgvector Redis AWS EC2 Docker Traefik Cloudflare Linux Git CI/CD model deployment LLM deployment monitoring fine-tuning pipelines OCR fine-tuning inference optimisation local hosting Ollama vLLM Unsloth multi-tenant architecture Copilot Studio
Report this Profile?