Agentic AI Engineer

AI Agent / Retrieval Engineer — Delhi
Company Description
Moleculyst Ventures is building autonomous AI systems capable of searching, reasoning over, and working with large real-world information systems. We are currently developing agent architectures that can independently explore large databases, retrieve relevant information, maintain useful memory, and perform multi-step research and engineering tasks.
Role Description
We are looking for an AI engineer in Delhi who can work closely with us on building autonomous agents and semantic retrieval systems.
One immediate problem is enabling an LLM agent to intelligently explore a large existing database. Traditional indexing can efficiently retrieve exact or structured matches, but an agent often needs information that is semantically related to its current question. We want to build retrieval infrastructure that allows an agent to discover this information reliably while maintaining enough observability and evaluation to determine whether the agent is actually searching effectively rather than merely producing plausible answers.
This is not primarily a prompt-engineering role. You will work on the underlying systems: database exploration, semantic indexing/search, embeddings, retrieval, agent tool use, memory, evaluation, and potentially knowledge representations such as knowledge graphs where they are genuinely useful.
What you'll work on
Semantic search and indexing over large existing databases
LLM agents capable of autonomously inspecting schemas and querying data
Retrieval pipelines using embeddings/vector similarity where appropriate
SQL/Python integration and database tooling
Agent memory and knowledge representation
Evaluations for retrieval coverage, precision, recall, and agent reliability
Experiments comparing approaches such as vector retrieval, hybrid retrieval, structured search and knowledge graphs
Building systems from first principles rather than depending entirely on agent frameworks
Who we're looking for
Strong Python and software-engineering fundamentals
Comfortable with SQL and databases
Understanding of embeddings, transformers and modern LLM systems
Ability to understand unfamiliar technical areas very quickly
Comfortable reading papers/documentation and turning ideas into experiments
Strong debugging and experimental discipline
Able to challenge an architecture when evidence suggests a better approach
Available to work in person in Delhi for substantial portions of the project
Experience with FAISS, vector databases, information retrieval, RAG, knowledge graphs, search systems or agent architectures is useful, but not mandatory if you can learn these quickly.

Work arrangement
Hybrid