I'm a 2nd-year CS student specialising in AI/ML, building things that (hopefully) matter.
So far: a CV system for rural child health screening in India, an LLM agent that scored 0.80 on Medium difficulty at Meta's AI agent competition, and a CNN that taught me more about augmentation tradeoffs than about plant seedlings.
I care about systems that work in the real world — not just notebooks that pass. I'm most interested in LLMs, computer vision, and the gap between a model that works and one that ships.
Currently looking for an AI/ML or Data Engineering internship where I can build something real and learn fast.
- 🧪 Wrapping my House Price model in a Flask API — making it actually deployable, not just a notebook
- 📖 Going deeper on LLM evaluation frameworks — prompted by what I saw scoring at Meta
- 🛠 Improving the Vision Shield pipeline's accuracy and edge-case handling
Core ML / DL
LLMs & Deployment
Computer Vision
Infra & Data
| Project | What it does | Result | Deploy |
|---|---|---|---|
| 🤖 Expense Audit Agent | LLM agent auditing employee expenses against company policy — built for Meta's AI Agent Hackathon (2025). Owned inference.py, Dockerfile, and full Hugging Face Spaces deployment via Docker + Groq API (Llama 3.1). |
0.80 Medium · 0.78 Hard · 0.67 Easy | |
| 👁 Vision Child Wellness Shield | Non-invasive BMI estimation from live camera feeds using MediaPipe Pose + pixel-to-cm calibration. Built for rural health screening under India's Mid-Day Meal Scheme. Sole developer on cv-development branch. |
Height ±5–10% · Weight ±10–15% · 15 images | |
| 🌱 Plant Seedling Classification | 3-block CNN classifying 12 species across ~4,750 images. Deliberately studied augmentation tradeoffs — found conditions where augmentation hurt generalisation. | ~70% baseline accuracy | |
| 🏠 House Price Prediction | End-to-end ML pipeline on 1,460 samples · 81 features. Deliberate imputation strategy + top-6 feature selection. Flask API wrapper in progress. | R² ≈ 0.79 |
