Agentic AI Engineer

Role

As our Agentic AI Engineer, you will build the intelligence layer that sits on top of our knowledge graph: the part of APAS that turns structured regulatory and operational data into deterministic, citation-backed answers and, over time, autonomous agentic workflows. This is a specialist role. We need someone who lives in large language models, retrieval architectures, and agent design, not a generalist ML engineer picking this up as one of several things they do.

Every model you build depends on data structured below you, and every decision you make about grounding, hallucination control, and citation accuracy has to hold up to the same standard the rest of APAS is built on: deterministic, defensible, provable.

This is not a job you'll do in a silo. Expect tight and transparent communication loops across our team.

Must Have

Minimum 5 years of hands-on experience building and shipping production ML/AI systems, including at least 3 years focused specifically on LLMs, NLP, or generative AI

Minimum 3 years of production Python experience, with frameworks like PyTorch, LangChain, or Hugging Face Transformers

Deep, practical experience with retrieval-augmented generation (RAG) architectures, including chunking strategies, embeddings, vector search, and hybrid retrieval that combines structured and unstructured data

Proven experience building agentic systems: multi-step reasoning, tool use, task planning, and orchestration (LangGraph, LlamaIndex, AutoGen, or a custom-built framework)

Strong grounding in prompt engineering and techniques for hallucination control, citation accuracy, and deterministic output, especially in high-stakes or regulated domains

Experience integrating LLM applications with knowledge graphs, ontologies, or structured data sources such as SPARQL/RDF. You don't need to own the graph, but you need to work fluently with it

Experience working with function-calling and tool-use models — open-source (e.g., Hermes, and other fine-tuned Llama/Mistral variants built for agentic use) as well as proprietary (e.g., Claude, GPT) — and knowing when a self-hosted, fine-tuned model is the right call versus a foundation model API

Experience fine-tuning or adapting open-source or foundation LLMs for domain-specific applications

Experience designing evaluation frameworks for LLM output quality, accuracy, and safety, not just spot-checking outputs by hand

Comfortable working with cloud infrastructure (AWS or Azure) and collaborating with DevOps on deployment, even though you won't own infrastructure yourself

Strong communicator who can explain model behavior and trade-offs clearly to non-technical stakeholders, including our go-to-market and leadership team

Nice to Have

Experience in regulated, compliance-heavy, or government and public-sector domains, where deterministic and defensible AI output genuinely matters

Familiarity with knowledge graph query languages (SPARQL, Cypher) beyond a surface level

Experience building digital-twin or simulation-style AI features

Prior startup experience, especially pre-revenue or early-stage, and comfort building from zero with a high degree of ambiguity

Open-source contributions to LLM tooling, agent frameworks, or RAG libraries

Exposure to the water utility, civil engineering, or environmental regulatory space

Location

India, Remote

Compensation

$1,500 - $2,500