I build things at the intersection of AI and the web — from fine-tuning large language models to shipping full-stack applications that make those models actually useful. Currently deep in the internals of LLMs: how they work, how to make them efficient, and how to make them explainable.
Third-year CS undergrad Geethanjali College of Engineering and Technology · AI & ML · 2027 Batch
- ex-LRP-Turbo — a custom LLM merging algorithm built on Arcee AI's Mergekit that combines three ideas: explainability (understanding what a model has learned), AttnLRP (layer-wise relevance propagation for transformers, to score which layers matter most), and TurboQuant (Google's near-optimal vector quantization, for efficient merging). The core idea: take a base model (global) and a fine-tuned version (local), use LRP relevance scores to decide which layers to keep from each, and apply TurboQuant to compress without losing what matters.
Applications: Medical AI (merge a general clinical model with a specialist fine-tune while keeping only the layers that learned domain-specific reasoning), edge deployment (produce smaller merged models that retain task performance without full fine-tuning costs), federated learning (merge locally fine-tuned models from different clients into a global model intelligently rather than averaging weights blindly), and fake news / misinformation detection at scale (the original use case — lightweight models that can be deployed without GPU infrastructure). More details: PR #682
- Building my project portfolio — AI/ML and web dev — for 2026 and 2027 campus placements and internship applications
- LLM fine-tuning — LoRA, QLoRA
- MLOps & model deployment (getting my models out of Colab)
- Full-stack development with AI-powered applications
- Open source AI/ML projects — NLP, LLM tooling, model merging and fine-tuning
- Full-stack web development projects, especially ones with an AI layer on top
- Anything where the goal is understanding, not just benchmarking
- Honest feedback on my projects — research rigor, problem solving-oriented and real-world applicability
- Connections in the AI/ML and software industry
- Building a resume that gets noticed and preparing for campus placements as a CS undergrad
