Faizan Mohammed Jamal Shaikh

LangGraph, Supabase (pgvector), sentence-transformers(BAAI), NVIDIA NIM, Python

• Architected a 6-node agentic RAG system in LangGraph (retrieve, grade, rewrite, generate, verify, fallback) that grades and verifies grounding before answering, cutting hallucinations versus naive retrieve-then-generate RAG.
• Built a retrieval layer on Supabase pgvector with a 384-dimensional HNSW index and local sentence-transformer embeddings, returning the top-6 ranked chunks per query with inline citations via the NVIDIA NIM API.
• Implemented automatic query rewriting (up to 3 retries) and an LLM fallback path, keeping the system able to answer even when the document corpus lacks direct coverage.

• Designing a specialized "expert model" by fine-tuning a compact 3B parameter/weights model Llama-3B on curated open-source repositories and licensing metadata to auto-architect zero cost product tech stacks, matching 70B model recommendation quality.
• Implemented QLoRA & DPO Alignment via Unsloth and Hugging Face TRL to minimize architectural hallucination and enforce free/permissive-tier constraint compliance.
• Synthesized Synthetic Instruction Pair s mapping high-level software requirements to optimal open-source tools, framework tradeoffs, and deployment paths within hardware budget limits.

React, Supabase, Stripe, Computer Vision

• Built a privacy-first image redaction web app in 24 hours at the NJx Hackathon, acquiring 50+ active users.
• Engineered a client-side pipeline to scrub EXIF metadata and redact PII using computer vision, ensuring zero server-side data persistence for 100+ images processed. Integrated Stripe for usage-based billing and Supabase for secure backend authentication.

Next.js, React, Python, Docker, Electron

• Engineered a desktop IDE application by forking VS Code with Electron and Next.js, enabling multi-developer real-time collaboration on unified software projects.
• Containerized workspace environments using Docker, ensuring synchronized package dependencies across all collaborators. Integrated a multi-agent AI orchestration engine that pools individuals AI agents to optimize context sharing and reduce overall LLM token usage.

Lead Developer

• Developing a secure interview environment that actively prevents cheating by blocking unauthorized AI assistants (like Claude) and external resources during assessments.
• Implementing advanced monitoring capabilities to ensure interview integrity while maintaining a seamless candidate experience.

Co-founder & Developer

• Building Thozha, a personalized settlement platform that brings everything an international student needs into one place, addressing scattered resources, lack of personalization, and weak institutional support.
• Developing core features including real-time housing listings, automated banking setup, curated news, structured networking, a secure marketplace, and visa-specific document guides.
• Creating a real-time, actionable dashboard tailored to student's visa type, country of origin, budget, and timeline to help over 1.1 million international students transition to the U.S. smoothly.

Founder & Lead Developer

• Architecting a full-stack social platform (React, Node.js) connecting students and developers for project collaboration and talent matching.
• Implemented user matching, squad formation, and real-time project discovery; targeting 100+ active users by Summer 2026.

• Designed and deployed a full-stack puzzle game featuring leaderboards, user preferences, and high-score tracking.
• Built a production application serving global users with a responsive frontend and backend data persistence.

Undergraduate Research · NJIT

• Benchmarking LLMs on 400+ SNOMED CT concepts to evaluate their performance as ontology servers.
• Building evaluation pipelines with Python, Pandas, and REST APIs in collaboration with a PhD student and faculty mentor.