Experience
6 - 11 yrs
Job Location
India
Vacancy
1
Designation
AI Engineer
Job Type
Not specified
Job Description
About the role
Our Agents service line builds custom agents on Anthropic Claude and OpenAI direct not a chatbot wrapper, not a per-resolution rebrand. We replace per-resolution agent SaaS (Salesforce Agentforce, Intercom Fin, Zendesk Resolve) with custom agents that run on the client s own LLM contract. AI Engineers here own the agent architecture, eval harnesses, and production stability.
What you'll do
- Architect agent loops: planner / executor / verifier patterns, tool-use schemas, retry + escalation logic.
- Build retrieval pipelines: chunking strategy, vector DB choice (Pinecone / pgvector / Weaviate), reranking, freshness handling.
- Own eval harnesses: golden-set construction, regression dashboards, cost / latency / accuracy SLAs.
- Stay current on Anthropic + OpenAI release notes; adjust prompts + schemas when APIs evolve. We treat the foundation-model contract as a dependency, not a vendor.
- Ship to the client s AWS + Anthropic / OpenAI accounts. Their tokens, their data, their audit log.
Must-haves
- 6+ years building production software. At least 1 year shipping LLM-powered features to real users (not POCs).
- Strong Python or TypeScript. Comfortable writing structured-output schemas (Pydantic, Zod) and reading Anthropic + OpenAI docs as primary references.
- Has built and maintained an eval harness with regression detection. Knows the difference between vibes-eval and a real harness.
- Comfortable cost-modelling agent loops in tokens. Can predict spend on a 100K-resolution / month workload before deploying.
Nice to have
- Open-source contributions to LangChain, LlamaIndex, Anthropic SDK, Instructor, or DSPy.
- Public writing on agent / RAG architecture.
- Experience with regulated-industry deployments (legal, healthcare, financial).