AI / ML Engineer

ITLabz Technology; India; On-site; INR 1,800,000 - 3,400,000; Full time

Job summary

ITLabz Technology is seeking an ML/software engineer to build production RAG and agent systems on Azure OpenAI and open models. You will own retrieval quality end-to-end, design guardrails, and instrument cost and latency, with a practical focus on datasets, embeddings, and evaluation.

The role requires three+ years in ML or software, strong Python, and experience running end-to-end LLM applications. This is an opportunity to ship impactful AI that handles real users.

Qualifications

• 3+ years in ML or software with production LLM systems.
• Strong Python and async/streaming for LLM apps.
• Experience with embeddings and vector search.
• Experience running evaluations and learning from results.

Responsibilities

• Build production RAG and agent systems on Azure OpenAI and open models.
• Own retrieval quality end-to-end: chunking, ranking, grounding, citations.
• Build evaluation suites with fixed questions and known-good answers.
• Design guardrails against prompt injection and data leakage.
• Instrument cost and latency per request and communicate trade-offs.

Skills

Python

LLM applications

Vector search

RAG systems

Async programming

Tools

Azure OpenAI

Anthropic

Open models

Job description

Ship production RAG and ML systems on Azure OpenAI and open models.

You will build AI that survives contact with real users — which mostly means retrieval that finds the right thing, evaluation that catches regressions, and knowing when the honest answer is that a language model is the wrong tool. We ship RAG assistants, agents and document intelligence into production, with citations and refusal behaviour rather than demos.

What you’ll do

• Build production RAG and agent systems on Azure OpenAI, Anthropic and open models
• Own retrieval quality end to end: chunking, ranking, grounding and citation correctness
• Build the evaluation suites — a fixed question set with known-good answers, scored on every change
• Design guardrails against prompt injection, data leakage and confidently wrong answers
• Instrument cost and latency per request, and make the trade-offs visible to the client
• Say plainly when a problem is better solved with a query, a rule or a classifier than with an LLM

What we’re looking for

• Three or more years in ML or software engineering, with production LLM systems you can describe in detail
• Strong Python, and comfort with the async and streaming shape of LLM applications
• Practical grasp of embeddings, vector search and where retrieval actually breaks
• You have run an evaluation and changed your mind because of it

Nice to have

• Fine-tuning or preference tuning in production
• Document intelligence over messy PDFs and scans