CS student at Islington College, Kathmandu. I build ML models and the products around them — PyTorch, RAG pipelines, agent tool-calling, and full-stack deployment.
CS student in Kathmandu. I got into programming out of curiosity and ended up falling for machine learning — first the fundamentals (PyTorch, training models from scratch), then LLMs and RAG, and now the part I enjoy most: building AI products around models, not just notebooks.
- 🔭 Currently building an AI research agent — a custom agent loop (no frameworks) that combines document RAG with live web search.
- 🌱 Learning deep learning + LLM engineering in parallel. I learn by building — every project exists because I wanted to understand one specific thing.
- 📫 Reach me at Praansu12@gmail.com — always happy to chat.
|
An agent loop built from scratch — no framework. Decides on its own whether to search your documents (ChromaDB RAG), search the web, or both. Streams tool calls live.
|
Upload a PDF, ask questions about it. Full RAG pipeline — PyMuPDF extraction, sentence-transformers embeddings, ChromaDB retrieval, Groq Llama 3. Streaming SSE + document deletion.
|
|
PyTorch + ResNet18 transfer learning — classifies bus/car/motorcycle/truck at 90% accuracy (400 images, 100/class). Confusion matrix visualization. FastAPI inference API with Docker.
|
Full-stack e-commerce for eco products — Next.js 16, TypeScript, Tailwind, Prisma, Stripe. Blur-up image placeholders, debounced search with URL sync, cart persistence.
|
| Project | What it is |
|---|---|
| IoT smart parking — ESP32 firmware, FastAPI backend, live dashboard | |
| ❤️ Health Guard ML | Heart disease prediction with XGBoost + SHAP explanations |
| 🏷️ Vehicle Labeling Tool | PyQt5 GUI I built to label my own training data |
| 🔬 Small Agent Reliability | Research: evaluating 3B-9B models as autonomous agents across 14 tasks |
Languages — Python, TypeScript, JavaScript, HTML/CSS, SQL, Java, C++
ML/AI — PyTorch, scikit-learn, Pandas, NumPy, XGBoost, OpenCV, sentence-transformers
LLM / RAG — Groq (Llama 3), ChromaDB, embeddings, agent tool-calling, RAG pipelines
Backend & Tools — FastAPI, Next.js, Prisma, Streamlit, Flask, Git, Docker, SQLite
IoT — ESP32, Arduino