I've built payments, marketplaces and ledgers.Now I build the LLM systems that sit on top of them.Because a clever prompt never survives a bad architecture.
Senior Software Engineer working on production Generative AI, agentic pipelines and enterprise blockchain for a US company, remotely from Kathmandu. Polyglot by necessity: Go for services, TypeScript for products, Python for models, PHP where the money still lives. I care about retrieval quality, evaluation and boring, observable infrastructure more than about the model of the week.
Track record
2023 — Pres.
Senior Software Engineer · Remote
Production LLM applications and Retrieval-Augmented Generation for enterprise data. Advanced RAG pipelines, agentic workflows and an evaluation framework so we know when retrieval gets worse, not just when it gets slower. Go services and Hyperledger Fabric chaincode underneath.
2022 — 2023
Full Stack JavaScript Developer · Hong Kong
End-to-end delivery on the MERN stack for client products, from MongoDB data modelling to React front-ends. Earlier in the same company: Laravel APIs with Vue.js front-ends.
2019 — 2022
Software Engineer · Kathmandu
Fintech web applications and internal tooling for one of Nepal's largest digital payment providers. Grew from Associate to Software Engineer while shipping on payment and banking products where a bug is a customer's money.
2019
Full Stack Laravel Developer · Kathmandu
Full-stack web applications with Laravel and MySQL, taken from Adobe XD mock-ups to deployment.
Principles
An LLM feature without an eval set is a demo. I build the faithfulness and recall checks first, then the feature, then the dashboard that tells us when it drifts.
Most 'the model hallucinated' bugs are 'we retrieved the wrong chunk' bugs. Chunking, hybrid search and reranking get more of my attention than prompts.
Go services, queues, idempotent handlers, structured logs. The exciting part of an AI system should be the answers, not the incident channel.
Ingestion, retrieval, orchestration, evaluation, deployment. Splitting them across five people is how you end up with a pipeline nobody understands.
Toolbox
Side projects
Contact
Senior roles, contract work, or a second opinion on a RAG pipeline that “mostly works”. One email, no form, no calendar link.
santoshbhulun@gmail.comReplies within a day or two. Your data stays between us.