Web developer. Now a AI engineer.
I'm Manish Karki. I spent five years shipping WordPress sites and Python web apps for real businesses. Today I'm an AI engineering intern, learning to take large language models from research to production.
Five years, in numbers
What I'm working on
Building RAG that doesn't hallucinate
- Shipping a retrieval-augmented chatbot for an internal knowledge base at my internship — hybrid search, reranking, citation enforcement.
- Writing an eval harness so we can actually tell when a prompt change makes things better or worse, instead of vibes.
- Studying transformer internals end-to-end — attention, KV caching, positional embeddings — so the abstractions stop feeling like magic.
- Reading Designing ML Systems by Chip Huyen and The Hundred-Page ML Book.
Selected projects
docu-chat
A grounded chatbot over PDFs and markdown with hybrid search, reranking, and forced citations. The project that taught me the full RAG loop.
- Python
- FastAPI
- LangChain
- Chroma
prompt-bench
A tiny evaluation harness that scores prompt variants across a fixed test set on accuracy, cost, and latency. Built during my internship, used daily.
- Python
- Pytest
- Streamlit
neural-from-scratch
A tiny neural network library in pure NumPy — forward pass, backprop, optimizers. I built it to demystify what PyTorch is doing under the hood.
- NumPy
- Jupyter
- Matplotlib
Recent writing
The RAG mistake I made three times before I caught it
Chunking by tokens, not by meaning. Retrieved snippets that ended mid-sentence kept poisoning the answer.
Why I'm still glad I started in WordPress
Shipping to non-technical clients teaches you things about software that no bootcamp will.
Eval-driven prompting: a tiny case for taking it seriously
Without a test set, you're not iterating on a prompt — you're rearranging deck chairs.
I'm looking for an AI engineering role where I can keep growing.
Junior, associate, or a meaty internship — if you're building things with LLMs and you need someone who already knows how to ship, I'd love to hear about it.