Technical partner for founders building with AI
Make AI earn its place.
I find manual work worth automating, reduce avoidable API spend, and build production AI systems your team can understand and own.
Evidence, not adjectives
01
Model judgment when AI matters. Systems judgment when everything around it matters more.
I work across research, application code, infrastructure, cost, and operations. Founders get one technical partner who can decide where AI belongs, build the path, and leave behind a system instead of a demo.
- 01Automation opportunity mapping
- 02AI pipeline architecture and integration
- 03RAG, language, and speech systems
- 04API cost and production reliability
02
Work where a wrong answer has consequences.
Public research, operated automation, and sanitized production systems. Every claim carries a result, boundary, or artifact.
91% Precision@1
Legal retrieval for a language mainstream benchmarks overlook.
Grounded answers across 10,320 Nepali case laws, with the simple baseline measured before architecture became complicated.
Full-history audit16.7M
movements checked across five years
Three independent tests · zero unexplained breaks16,763,422 movements
Proving every movement in a five-year ledger still held together.
Three independent integrity tests, one corrected measurement error, and zero unexplained breaks. Confidential system, public engineering account.
365 companies tracked
A market-data system designed to keep operating without babysitting.
Scheduled collection, versioned data, public documentation, and no permanent application server for Nepal's listed market.
03
Start with economics. Build only after the system makes sense.
Find leverage
Map workflow, model fit, API spend, integration boundaries, operational risk, and expected value.
Own the path
Implement model, application, deployment, observability, and guardrails as one production system.
Remove dependency
Document decisions, expose operating cost, train your team, and leave rollback paths behind.
Have a manual workflow, expensive AI pipeline, or prototype that must become reliable?