4 years in production. KYC/AML compliance systems, fraud detection, multi-agent LLM pipelines — built and shipped in regulated financial environments.
- ML systems end-to-end — data pipelines → model training → inference APIs → monitoring → drift detection
- LLM & agent engineering — LangGraph multi-agent workflows, RAG pipelines, self-correcting agentic loops
- ML platform layer — A/B testing, experiment tracking, drift detection, Terraform IaC, CI/CD
- Compliance AI — AML screening, KYC risk scoring, SHAP-explainable predictions, regulatory reporting automation
Domain: KYC · AML · FATF · Insurance · Fintech · GDPR Stack: Python · PyTorch · LangGraph · FastAPI · Docker · MLflow · Terraform · AWS
FT-Transformer + anomaly head for real-time transaction fraud detection. AUPRC 0.9676 · AUROC 0.9947 · F1 0.9439 Full MLOps: MLflow · FastAPI · Docker · GitHub Actions CI/CD · Streamlit dashboard
Ensemble AML scoring system calibrated on the Tunisian BCT/CTAF regulatory framework. 49 features · 10 CTAF typologies · XGBoost + LightGBM + Isolation Forest ensemble · SHAP explanations (auditability requirement) Feeds the Waraka STR drafting pipeline upstream.
LangGraph agent that takes plain French descriptions of suspicious transactions and produces goAML-compatible STR XML for CTAF submission — under BCT circular n°2025-17. Dual-sector: banking + insurance. Stack: Claude Sonnet · OpenSanctions · FastAPI · PostgreSQL · Redis · ChromaDB · 54 tests passing
Multi-model ML platform: recommendation, fraud detection, demand forecasting, ticket classification. Platform layer: deterministic A/B testing · KS + PSI drift detection · MLflow tracking Infrastructure: FastAPI · Docker Compose · MLflow · Terraform on AWS (EC2 + S3)
LangGraph multi-agent pipeline: parallel analysis agents (asyncio) + self-correcting improvement loop (Critique → Improve → Adversarial Verifier). SQLite checkpointing · React/Vite frontend · FastAPI
RAG API for insurance policy Q&A: multi-stage retrieval, category-aware intent detection, hybrid reranking. Qdrant · Ollama · FastAPI
| System | Result |
|---|---|
| AML entity matching + risk scoring | ~30% reduction in false-positive compliance flags |
| ETL pipelines over 100K+ client records | ~40% reduction in manual compliance work |
| OCR model for document ingestion | ~60% reduction in manual data entry |
| RPA automation bots | ~20% operational capacity recovered |
AI / LLMs: Python · PyTorch · LangGraph · LangChain · RAG · GraphRAG · Hugging Face · OpenAI · Groq ML / MLOps: scikit-learn · XGBoost · LightGBM · MLflow · GitHub Actions · pytest · NLP · OCR · SHAP Platform / Infra: Docker · Terraform · FastAPI · A/B Testing · Drift Detection Vector DBs: Pinecone · ChromaDB · Qdrant · pgvector Data / Backend: PostgreSQL · MySQL · MongoDB · REST APIs Domain: KYC · AML · FATF · Financial Risk · Insurance · GDPR Languages: French C2 · English C2 · German A1 (active)
LinkedIn · Kaggle · achrafgasmi58@gmail.com Open to relocation — Netherlands · Germany · France · Belgium · Switzerland · Luxembourg