Jamil Shah

I’m a Software Engineer with 3+ years of experience building production Generative AI systems for enterprise document intelligence. I work across the full stack of the problem: LLMs, RAG pipelines, hybrid retrieval (BM25 + vector search), and prompt engineering, paired with OCR and computer vision to pull structured data out of messy, real-world documents.

At CogniQuest Technologies, I led development of a distributed AI-driven invoice processing system for Flipkart India. I architected pipelines that extract financial data such as GSTIN, invoice values, and metadata using LLMs, OCR, and NLP models, reaching 85% extraction accuracy. By combining ML with rule-based validation, I cut manual validation effort by 60%. The system processes 10,000+ documents a day with sub-2s average latency, built on Celery, Redis, Docker, and microservices. I also built the table parsing, financial computation, and cross-document validation engines that keep the extracted data consistent at scale.

Outside client work, I’ve built a multi-tenant RAG Q&A platform that improved answer relevance by 35%, a modular document classification framework with a model registry and zero-downtime rollback, and an email automation system that reduced manual triage by 45%.

My stack is Python, FastAPI, Django, PostgreSQL, RabbitMQ, GCP, and Docker. I enjoy owning problems end-to-end, from architecture to production, and turning ambiguous, document-heavy workflows into reliable, measurable AI products.