AI Engineer with professional experience developing LLM-based software analysis systems, multi-agent workflows, RAG pipelines, and FastAPI backend services. Skilled in building AI systems for code analysis, repository metadata processing, retrieval-augmented generation, query routing, reranking, guardrails, caching, and Dockerized deployment. Experienced in applying LLMs and automation to improve software vulnerability analysis, document-based decision support, enterprise knowledge retrieval, conversational workflows, and deep learning applications.
Fine-tuned Llama 3.1 8B (QLoRA + Unsloth) on 7,000 synthetic examples, achieving 1.00 citation recall and 96.6% JSON accuracy, deployed via self-hosted vLLM for privacy-preserving policy assistance. Built an adaptive LangGraph RAG pipeline (FastAPI, hybrid Qdrant retrieval, cross-encoder reranking, corrective Self-RAG, fail-closed answer verification, Redis caching) achieving 1.00 RAGAS faithfulness and context recall.
AI-powered claims platform that reduces slow, manual insurance review by extracting hospital invoice data, auto-filling claim forms, and generating evidence-backed approval, rejection, or human-review reports. Built with FastAPI, LangChain, FAISS, FlashRank, OCR pipelines, Redis caching, and Docker, implementing hybrid BM25 + FAISS retrieval, RRF fusion, reranking, and confidence-gated Self-RAG checks.
AI-powered Banglish chatbot platform that reduces the hassle of traditional bus ticket booking by helping Bangladeshi users search, compare, and book tickets through natural conversation. Built with FastAPI, SQLite, Gemini, LangChain, ChromaDB, and Docker, implementing RAG with MMR retrieval, query rewriting, exact route lookup, and provider policy enrichment for route, fare, schedule, and policy queries.
Open to AI engineering roles, research collaborations, and meaningful real-world AI problems.