One project at a time
I take on the occasional project
Not many, and only where the problem is one of these four.
I am in my final year of computer science at NIT Bhopal and most of my time goes into the product I am building, so I am not looking for volume. But if what you have is genuinely one of the things below, I am interested, and I would rather you write than assume I am busy.
What fits
- Retrieval that is measurably wrong
- A search or RAG system that works on the sample data and returns the wrong thing in production, where nobody can say which layer is at fault. I build an eval set from real traces first, because the argument is usually about whether it is even broken.
- AI that has to run on the user's machine
- On-device inference that has to feel instant rather than merely private. Apple Silicon, MLX, Core ML, Metal.
- Real-time media where latency is the hard part
- Video encoding, streaming and transport. QUIC, WebTransport, WebCodecs, hardware encoders.
- Interfaces that are supposed to look impossible
- Shaders, particle systems, and the kind of WebGL work that has to hold sixteen milliseconds.
What does not
Prompt tuning as a service, AI strategy decks, or a vector database migrated with no evaluation attached. Someone cheaper will do those just as well, and I would only be slower at them.
I will also tell you honestly if your timeline and mine do not fit, rather than taking it and being late.
Get in touch
Tell me what you are building and where it is stuck. If I am not the right person I will say so, and I usually know who is.
Hiring rather than contracting? That is a shorter conversation. I graduate in May 2027.