You will work in a small team on Python services, LLM agents (Google ADK on Vertex AI), evaluation, and the review UI that analysts and actuaries use every day.
Role details
Title
Junior AI Engineer
Experience
2–5 years
Team
Epsilon Labs
Location / work mode
Bangalore
You will ship features end to end, from a written spec to tested code, with a senior engineer reviewing your work.
• Build and improve LLM agent steps (mapping, extraction, review) using Google ADK and Gemini on Vertex AI, with versioned prompts and strict JSON output schemas.
• Write the deterministic code around each agent: input preparation, citation and quote verification, and a code stand-in with the same contract.
• Parse documents (DOCX, PDF, XLSX) into sections, tables and values; work with calculation workbooks and their formulas.
• Extend Pydantic data models and FastAPI endpoints, and keep generated JSON schemas current.
• Measure quality: build evaluation sets from analyst decisions, track recall and precision, and record run cost and latency.
• Write pytest tests that run offline with no credentials, using recorded model responses (cassettes).
• Contribute to the analyst review UI (HTML, CSS, vanilla JavaScript).
• Follow the spec-driven workflow: read the spec, plan, implement tasks, and keep code traceable to the spec section it serves.
• Help with persistence, telemetry, audit logging and cloud deployment (GCP and AWS) as the product moves toward customer release.
• Python 3.11+
• LLM APIs: prompting, structured JSON output
• Pydantic, JSON Schema
• pytest
• REST APIs (FastAPI or similar)
• Git and pull requests
• Document parsing: DOCX, PDF, XLSX
• Evaluating model output
• Clear written English
• Degree in CS, Engineering, Maths or Statistics, or equivalent experience
• Google ADK, Vertex AI, Gemini
• Agents: tool calling, multi-agent flows
• RAG, embeddings, retrieval
• LLM evaluation: golden sets, precision/recall
• Excel formulas, openpyxl, pycel
• Insurance or actuarial domain
• SQL (PostgreSQL, SQLite)
• HTML, CSS, JavaScript
• Azure, GCP or AWS, Docker, CI/CD
• Logging, metrics, tracing
• Authentication, RBAC, roles, audit logging
• Spec-driven development
• AI coding assistants
• Prompt Engineering