Talal Muzaffar

Hey! I'm Talal, a machine learning engineer working on document AI, retrieval, and agentic systems — currently in Barcelona, moving to Paris this September.

ml lead at DOST AI

architected RAG @ 3.8M financial docs AskDOST · agentic document Q&A extraction · compliance · discrepancy detection

built with PyTorch LangChain LangGraph Google ADK MCP A2A FastAPI

recognised Google AI Booster — winning team, Madrid 2025

studying at EPITA Paris · MSc Data Science & Analytics 2026 — 2028

studied at FAST National University · Data Science Dean's List ’24–’25

I work in the parts of AI that don't make headlines — turning document chaos into queryable systems. At DOST AI I architected a RAG pipeline over 3.8M+ financial documents, then built the agentic layer on top: AskDOST, which answers natural-language questions about invoices and workflows through a streaming tool-use loop over MCP, with every write gated behind explicit user confirmation. Full projects list here.

These days I lead a five-person team owning that stack, across customers in ten countries. Most of the interesting work is delivery-shaped: InstanceBuilder collapsed a multi-day onboarding back-and-forth into one guided session, and vendor reconciliation that used to crawl now runs at 10+ enterprise customers a month.

I fine-tune when curious: the Falcon-7B experiment hit 0.90 BERT on MICCAI using 4-bit QLoRA on a single T4. Outside of that, I follow open-source closely and try to ship something every month.

I co-authored Choices for Measuring Cognitive Function Digitally in LMICs at the American Epilepsy Society, and was selected for McKinsey Forward and Harvard's Crossroads.

Find me on

GitHub LinkedIn Résumé (PDF)

Or mail me at talalmuzaffar3@gmail.com