Sahil Al Farib

I’m an AI/ML engineer (intern and final-year CSE student) who builds production-ready LLM/RAG systems and experiments that hold up in real deployments. I’m especially drawn to retrieval accuracy, hallucination reduction, and end-to-end pipelines that users can actually rely on.

In my internship, I owned and deployed a production RAG pipeline using LangChain, the OpenAI API, and Pinecone—covering chunking, vector indexing, and query routing end-to-end. I led prompt optimization and retrieval evaluation loops to reduce hallucinations and improve answer groundedness, then integrated the AI assistant into the product backend via REST APIs with React/Node.js surfaces.

On the research and open-source side, I published FactEval (a hallucination detection library for RAG) with calibrated confidence and claim-level diagnostics, and I built BanglaMixLoRA, a MoE-style Bangla language model using MixLoRA on Mistral-7B. I also created a real-time Bangla conversational AI avatar with sub-second latency via a multimodal ASR-to-generation-to-lip-sync streaming pipeline, and I’m actively working on few-shot cross-device quantum noise modeling on real quantum hardware.