Things I've built
that I'm willing to show you.
A mix of recent AI work, production Python web apps, and the WordPress sites that paid the rent. Each one has a case study with what worked, what didn't, and what I'd do differently.
docu-chat
A grounded retrieval-augmented chatbot that answers questions over a folder of PDFs and markdown. Built end-to-end to learn the full RAG loop: ingestion, chunking, hybrid search, reranking, and citation enforcement.
- Python
- FastAPI
- LangChain
- Chroma
- OpenAI
prompt-bench
A small evaluation harness for comparing prompt variants. Scores each on accuracy, cost, and p95 latency across a fixed test set and renders a report I can share with non-engineers. Used at my internship.
- Python
- Pytest
- SQLite
- Streamlit
neural-from-scratch
A tiny neural network library in pure NumPy — forward pass, backprop, optimizers, and a working MNIST trainer. I built it to demystify what PyTorch is doing under the hood.
- NumPy
- Jupyter
- Matplotlib
shelf — bookkeeping for solo founders
A Django app for a freelance client — invoices, expenses, month-end dashboard. In production for two years, used daily. The project that taught me how messy real-world data actually is.
- Django
- PostgreSQL
- HTMX
- Tailwind
kothi — boutique hotel theme
Custom WordPress theme + booking plugin for a chain of boutique hotels in Nepal. 100kb above-the-fold budget, Lighthouse 98+ on mobile. Still live, still earning the client revenue.
- PHP
- WordPress
- ACF
- Custom Gutenberg blocks