GOUD
I build agentic AI systems, multi-stage RAG pipelines, and local LLM deployments — end to end, from data collection to a working interface.
Built for depth,
not decoration.
B.Tech CSE (AI & ML) student focused on end-to-end AI systems — from data collection and model training to API deployment and user-facing interfaces. Most of my work runs on fully local LLM stacks: multi-agent RAG pipelines, self-evaluating retrieval loops, and the tooling to benchmark and deploy them without relying on paid APIs.
Roast Bot
Nothing special — but this is the project that pulled me into AI. Everything else here started with this.
THE PROBLEM —I wanted to see if I could get an LLM to talk back with actual personality, instead of a flat, polite response.
WHAT I BUILT —A small Streamlit app that feeds user input into a few-shot prompt telling Gemini to roast it, with a fallback to a personal API key when the shared free quota runs out.
Study Buddy
First time I built something that decides which tool to use, instead of just deciding what to say.
THE PROBLEM —Generic chatbots either make up answers or dump a whole document back at you instead of actually teaching from it.
WHAT I BUILT —An agent that reads a student's own uploaded notes and picks between a focused lookup tool or a broad summary tool depending on the question — and still works without an API key, using local embeddings.
Internet Researcher Agent
The first project where I stopped answering from memory and started building something that actually goes and checks.
THE PROBLEM —LLMs answer from training data, which is fine until you ask about something recent or specific.
WHAT I BUILT —A pipeline that searches the live web, embeds the results locally, drafts a sourced report, then keeps the follow-up conversation grounded strictly in that report instead of general knowledge.
Self-Improving RAG Agent
This is where I stopped babysitting the model's answers and let it check its own work.
THE PROBLEM —A RAG system's first answer is often mediocre, and most setups have no way to notice that or try again.
WHAT I BUILT —A retrieval loop that grades its own answer with a second model, and automatically retries with a wider search when the score is too low — running fully offline, at zero API cost.
Agentic RAG System
The project where I actually learned to make an agent say "I don't know" instead of guessing.
THE PROBLEM —Most RAG demos retrieve-then-answer no matter what — even for ambiguous questions or ones the documents just don't cover.
WHAT I BUILT —A router that chooses between clarifying, retrieving, calling a tool, answering directly, or refusing outright — reranking results before they ever reach the model, and refusing when the evidence doesn't support an answer.
Local LLM Performance Predictor
The project where I stopped guessing whether a model would fit on my machine and started measuring it.
THE PROBLEM —Downloading a big local model and hoping it doesn't choke your machine is a bad way to find out it won't fit.
WHAT I BUILT —A benchmarking harness that measured real latency, RAM, and VRAM across 8 local models, trained regression models on the results, and now predicts hardware risk through a small API and dashboard before you ever load the model.
What I build with.
Let's build
something.
Open to AI/ML roles and interesting agentic-systems problems. Reach out — I read everything myself.
DOWNLOAD RESUME