class Debjit:
def __init__(self):
self.role = "CS Undergrad @ IIT Jodhpur | Building Agentic AI Systems"
self.focus = ["LLM Orchestration", "RAG Pipelines", "AI Agents"]
self.currently_building = "Autonomous multi-agent pipelines that grade their own homework"
self.fun_fact = "I optimize latency the way some people optimize their morning coffee routine"
def say_hi(self):
return "let's build something that doesn't hallucinate 🤝"🔗 Link Hunter — AI Chrome Extension
A full-stack extension backed by FastAPI that analyzes web images and retrieves context-aware YouTube tutorials via a 2-stage multimodal pipeline (Gemma-4-31B-IT + Gemini embeddings + cosine re-ranking). A/B tested across 4 system prompts on a golden dataset, graded by an LLM-as-a-Judge.
| FastAPI LangSmith Pydantic Multimodal
📰 Multi-Agent Newsletter — Autonomous News Curator
A stateful DAG of 4 specialized LLM agents across 7 functional nodes, built on LangGraph. Implements an LLM-as-a-judge critic loop with a 3-retry limit and a hybrid RAG pipeline to keep every briefing grounded — zero context flooding, zero hallucinations.
LangGraph LangChain RAG Reflection Pattern
I test my agents like they're production systems — not demos.
Languages
AI / LLM Engineering
Vector DBs & Retrieval
Evaluation & Observability
Tools & Platforms
- 🎓 Top 3% — JEE Advanced'23 (AIR 8,772 / 2.5 lakh students)
- 🎓 Top 1% — JEE Mains'23 (AIR 14,772 / 12 lakh students)
