The role
Neon Link places remote AI engineers in India with US companies. Plenty of companies have tried a chatbot. Far fewer have an LLM application in production that answers from their own documents, calls their systems through tools and gets checked against real cases before every change ships. You build those: retrieval over a client's knowledge and records, agents that call tools to look things up and take the next step, and the evals that prove it works before a customer sees it.
We care more about what you built than the courses you took: show us something you shipped on an LLM API that real people used, tell us what it did and what kind of business ran it, and leave the client's name out. We are hiring for this role now. Depending on the engagement, the work is project based or full time, at a rate agreed up front. We never charge a fee to apply, test, train or start.
What you'll do
- Scope each LLM feature with the client: the question it answers or the task it does, the data it needs, and how you'll know it works
- Build production LLM applications in Python or TypeScript on the OpenAI, Anthropic or Gemini APIs, with structured outputs and model choices that fit the job's cost and speed
- Build retrieval (RAG) over the client's documents, tickets and records: chunking, embeddings, a vector store such as pgvector, Pinecone or Qdrant, and reranking where it helps
- Build agents that use tool calling to read from and write to the client's systems, with scoped permissions and a person approving anything that goes to a customer or moves money
- Write evals before you ship: a test set drawn from real cases, automated checks, and a regression run on every prompt or model change
- Ship it properly: logging, tracing, error handling, rate limits, and a cost and latency view the client can read
- Write a short runbook for each system so the client's team knows what it does, what it costs to run and who to call
What success looks like
- Answers that hold up against the eval set, with the quality bar and targets set with each client
- A system that keeps working when a model or API changes, because the evals catch regressions first
- A client team that uses it every day and can explain what it does without you in the room
What you bring
- 2 or more years shipping software in Python or TypeScript, including at least one LLM application that real users relied on; a project of your own counts
- Hands-on with retrieval: embeddings, vector search, chunking and the tradeoffs between them
- Built tool calling or agent workflows on the OpenAI, Anthropic or Gemini APIs, directly or through a framework such as LangGraph, LlamaIndex or the OpenAI Agents SDK
- You write evals and can show how you measured whether an LLM feature got better or worse
- Clear written English: you can explain a model's limits to a business owner who has never built one
- Careful with client data and access: you work only in the accounts a client sets up for you, keep their data, code and API keys out of personal accounts and public repos, and before your first client project you'll sign a confidentiality agreement, pass a background check and show a photo ID and proof of past work, such as experience letters, payslips or freelance invoices
- Education: any degree or none; what you shipped matters more
Nice to have
- Postgres and pgvector, Supabase, or a cloud platform such as AWS, GCP or Azure
- An eval or tracing tool such as LangSmith, Langfuse or Braintrust, or fine-tuning experience
Work setup
- Your own computer with 16 GB of RAM or more and an Intel Core i5, AMD Ryzen 5 or Apple M series processor or better, on Windows 10 or 11, macOS or Linux, with a webcam and a headset
- Broadband of at least 50 Mbps, with a mobile hotspot as backup
- UPS or inverter power backup so a power cut doesn't drop a client call or a deploy
- A quiet workspace for client calls and screen shares
- Tell us your city and state when you apply; we hire across regions, so one storm or outage doesn't hit everyone at once
You'll do well here if
Ships to real users
Measures before claiming wins
Plans for the failure case
Explains it plainly
Curious about the business
Careful with other people's data
Questions applicants ask
- Is this job work from home?
- Yes. Remote, work from home, anywhere in India.
- How much does it pay?
- INR 60,000 to 95,000 a month (about INR 357 to 565 an hour). About 7.2 to 11.4 LPA for a full-time client engagement, depending on experience. Project work before that is paid per project, at a rate agreed up front. Pay is in Indian rupees (INR).
- What are the working hours?
- Flexible, with at least 3 hours of overlap with US business hours (evening in India) for calls and handoffs.
- When would I start?
- Hiring now.
- How do I apply?
- Use the form on this page: your resume, a few questions about this role, your expected pay and a short voice introduction you record on the page. A person reads every application.
- Does Neon Link charge a fee to apply?
- No. Neon Link never charges a fee to apply, test, train or start, and only contacts applicants from neonlink.ai email addresses.