I build AI systems
and products that ship.
I’m Shishir Bhurtel. I build LLM agents, RAG pipelines and the full-stack apps around them.
- 01 /LLM agents & RAG
- 02 /Evals & tracing
- 03 /Backends & APIs
- 04 /Full-stack web apps
- Python
- FastAPI
- LangGraph
- LangChain
- RAG · pgvector
- OpenAI · Claude APIs
- Evals & tracing
- TypeScript
- Node.js
- NestJS
- Express
- Next.js
- React
- Golang
- Postgres
- MongoDB
- Redis
- Celery
- Docker
- AWS
Most AI features stall somewhere between the demo and production. I work in that gap — the AI layer and the system it runs on.
3+ years of industry experience and 2 years of freelance work — Node.js, NestJS and React first, now Python and FastAPI alongside them — with most of my time on LLM applications: LangGraph agents, retrieval over real customer data with pgvector, tool calling, Celery pipelines, and the evals and tracing that make them trustworthy. Today I build the AI layer at Spacebrain.ai; before that, 75+ projects for 40+ clients as a Top Rated freelancer. Based in Kathmandu, working with teams in the US and Europe.
- 3+ yrs
- Industry experience, plus 2 years of freelance work
- 75+
- Projects delivered for 40+ clients worldwide
- Top Rated
- On Upwork · Level 2 on Fiverr
- 100K+
- Concurrent users supported on a platform I re-architected
Things I’ve built.
AI products first, then the systems work underneath them. Open any project for the problem, the build and the links.
From first agent
to production.
Most engagements are a mix: an AI capability that has to be reliable, plus the system underneath it. Hire me for one step or the whole path.
Agents & LLM features
Assistants and agents that take real actions inside your product — drafting replies, preparing workflow steps, calling your APIs — with a human in the loop where it matters.
- Python · FastAPI services
- LangGraph · LangChain
- Tool calling & structured output
- Multi-step agent workflows
- OpenAI / Claude APIs
RAG & retrieval
Answers grounded in your own records and documents instead of the model's best guess.
- Retrieval over docs & customer records
- Embeddings & vector search
- pgvector · Postgres
- Python ingestion jobs · Celery
- Context & prompt design
Evals, tracing & guardrails
The unglamorous part: knowing when the model is wrong, what it costs, and whether a change actually made it better.
- Prompt design + evals
- LLM tracing & observability
- Guardrails for sensitive domains
- Cost & latency awareness
Backend, product & cloud
The system around the model — Python and FastAPI services, Node.js and NestJS APIs, data, interface, deployment. The difference between a demo and a service people rely on.
- Python · FastAPI · Celery
- Node.js · NestJS · Express · Fastify
- Golang · REST & WebSockets
- Next.js · React · TypeScript
- Postgres · MongoDB · Redis
- Docker · AWS · CI/CD
The ledger so far.
- 2025Oct 2025 — Present
AI & Backend Engineer
Spacebrain.aiBuilding the AI layer of a go-to-market platform that brings CRM, marketing, conversations and payments into one workspace. I work on the assistant and agent systems — LangGraph orchestration behind FastAPI services — plus retrieval over customer records and documents, tool-calling actions that draft replies and prepare workflow steps, and the evaluation and tracing that keep them trustworthy at scale.
Python · FastAPI · LangGraph · LLM orchestration · RAG · pgvector · Celery · Postgres · Docker
- 2024Dec 2024 — Jun 2025
Full Stack Developer (Contract)
AppCentric · United StatesWorking on the BackToIt and BidStruct products: optimising RESTful APIs, integrating third-party services and caching, and building the React and Redux interface for a government bidding platform.
Node.js · Express.js · React.js · Redux.js · MongoDB
- 2023Dec 2023 — Dec 2024
Full Stack Developer (Contract)
TijgerSoftware · GermanyDelivered national-level projects in the Netherlands and Germany, including a government exam assignment platform. Built APIs in Express and Next.js, OAuth and JWT authentication, Stripe payments, and improved the backend architecture to handle more than 100K concurrent users.
Express.js · Next.js · MongoDB · PayloadCMS · OAuth · Stripe
- 2021May 2021 — Jun 2023
Freelance Web Developer
Upwork · FiverrCompleted over 75 orders for more than 40 clients worldwide, reaching Level 2 on Fiverr and Top Rated on Upwork — mostly Node.js software, delivered directly with the client.
Node.js · JavaScript · React · MongoDB
- 01A comprehensive guide for PostgreSQL indexingApr 12, 2023 · 12 min
- 02Understanding the internal architecture of SlackApr 3, 2023 · 10 min
- 03Django vs Express — which framework should you choose?Mar 12, 2023 · 12 min
Patan Multiple Campus
2022 — 2026BSc in Computer Science and Information Technology (expected).
Trinity International College
2019 — 2020High school, GPA 3.64.
Open-source contributions, technical writing, and an unreasonable interest in how large systems stay up.
Questions
teams ask first.
What kind of AI work do you take on?
LLM features inside existing products, agents that call tools and take actions, RAG over your documents and records, and making an AI feature that already exists reliable enough to trust. I also build the full-stack system around it, so you don't need a second engineer to get it live.
Can you work inside our existing codebase and team?
Yes — that is most of what I've done. I've worked as a contract engineer inside product teams in the US, Germany and the Netherlands, and delivered 75+ projects directly with clients. I'm most productive in Python and FastAPI, Node.js with NestJS or Express, and Next.js with React.
How do you keep LLM features reliable in production?
Evals before and after every meaningful change, tracing on every model call, guardrails where the domain is sensitive, and a human in the loop for actions with consequences. I also watch cost and latency from the start, because a feature that works but is too slow or too expensive doesn't ship.
Where are you based, and how does the time zone work?
Kathmandu, Nepal (UTC+5:45). I've worked with teams in the US and Europe throughout my career — written updates by default, calls where they help.
How do we start?
Email me or use the form below with what you're building and where the model fits. I reply within a day. From there it's usually a short scoping call and a small first milestone, so you can judge the work before committing to more.