Founding AI Engineer

You'll work directly with the founders on the AI that sits at the core of Corpus: the agents, reasoning, retrieval, and voice that turn someone's financial life into advice they can trust and act on. This isn't a wrapper on top of an LLM; it's a system that reasons over portfolios, plans across steps, and executes real financial workflows. You'll own these systems end to end.

Responsibilities:

• Build the AI agents at the heart of the product's multi-step reasoning over a user's finances, planning, and executing financial workflows.
• Build retrieval and grounding over structured and unstructured financial data, so answers are accurate, current, and specific to the individual.
• Design the evaluation harness, guardrails, and observability measuring quality rigorously where being wrong carries real cost.
• Build the conversational experience, including voice, so talking to Corpus feels natural and fast.
• Build backend services, APIs, and data pipelines that take these systems from prototype to production.
• Stay ahead of a fast-moving model landscape and bring in what's worth adopting and discard what isn't.

Expectations:

• This is a high-ownership, in-person role in Bangalore. The work is research-adjacent but production-oriented; you'll prototype an approach in the morning and deploy it by evening. This works well for someone who's excited by frontier AI applied to a real, high-stakes domain. It may not be the right fit if you want to focus purely on research without shipping.

Requirements:

• 3-8 years building software, with depth in applied AI/ML, backend systems, or LLM applications.
• Strong Python and sound engineering fundamentals you write code others can build on.
• Experience taking LLM-powered systems to production: RAG, agents, and the unglamorous work of making them reliable.
• Be serious about evaluating; you measure model quality with rigor and let data drive decisions.
• Fluency with embeddings, vector stores, and orchestration frameworks and the judgement to know when not to use them.
• Fast, iterative operator: you ship, watch how real users respond, and improve quickly.

Bonus:

• Voice and speech models, or real-time conversational systems.
• Agent frameworks: LangGraph, LlamaIndex, CrewAI, or similar.
• Production vector stores (Pinecone, Weaviate, Qdrant, pgvector) and AI observability tooling.