Job description
Enterprises want agents that can do real work on their own data. That data sits across documents, databases and business systems that were never built for agents to read, and today's agents forget most of what they learn once a session ends.
Our team builds Sarvam's context engine to solve both problems. The knowledge side brings an enterprise's data together so agents can search it, navigate it and cite where an answer came from. The memory side lets agents remember what they learn over time, about individual users, about teams, and across the organisation. We want both to keep improving as they are used.
Knowledge and memory for agents is a rapidly evolving field. We want to build at the frontier of it, and to show with measurable results that our approach does better than traditional ones.
About Sarvam
Sarvam is India's leading full-stack AI company, building sovereign AI infrastructure and applications purpose-built for India. Headquartered in Bengaluru and founded by leading AI researchers, Sarvam builds foundational models for Indian languages, enterprise AI platforms, and voice-first AI agents that serve hundreds of millions of Indians. Sarvam is backed by top-tier investors and works with national and state governments, large enterprises, and developers across the country. We are building the AI layer for a billion people.
What this role promises you
You will join a small team and own large parts of the platform. The work is driven by real customer deployments, so what you build gets used quickly and you see what works and what does not. You will also work closely with the teams building Sarvam's models, document AI and inference, which makes it much easier to solve problems that cut across the stack.
What you'll do
- Build and run the ingestion pipeline that parses, extracts and indexes large volumes of documents, and keep it reliable when files are messy and external services fail.
- Connect to databases and enterprise systems so agents can work with structured records as well as documents.
- Design how enterprise data is modelled, including the ontology, matching the same entity across sources, and tracing every value back to where it came from.
- Build the retrieval tools agents call, with search, filters and citations, and make them fast and cheap enough for agents to use heavily.
- Build memory for agents at the level of a user, a team and the whole organisation: how it is written, stored, retrieved and kept accurate over time.
- Run evals, find where answers go wrong, and keep iterating until quality improves.
- Work with product and forward-deployed engineers to turn what one customer needs into something the platform does for every customer.
- Write design docs and keep decisions documented.
Our production stack is Go, with Temporal, OpenSearch and Postgres.
Who you are
We are hiring at 3 to 5 years of experience.
- You have built and run backend or data systems in production, and you have fixed them when they broke.
- You are strong in at least one backend language. Our core is in Go, so Go experience is a plus, and you should be happy to work in it every day.
- You have good fundamentals in data modelling and PostgreSQL.
- You have worked on long-running pipelines and know how to make them safe to retry and recover.
- You have worked on search or retrieval and know how to debug why the right result did not come back.
- You use AI coding tools every day and have a good sense of where they help and where they don't.
- You care about measuring quality, and you are comfortable building the evals to do it.
Bonus points
- Experience with knowledge graphs, ontologies, entity resolution or structured extraction.
- Experience building memory for agents or LLM applications.
- Document processing, such as OCR, layout or table extraction.
- Connectors or change data capture for ERP, CRM or other enterprise systems.
- Building on MCP, or serving LLMs and embedding models in production.
- On-prem or air-gapped deployments.
- Experience at an early- or growth-stage startup.
Posting text © Sarvam AI, shown from its public careers page for reference; the employer's page is the authoritative version. Posted 2026-10-23; listed here until 2026-12-22 unless the employer closes it earlier.
