SDE2 - Full Stack at sarvam, Bengaluru
Open. Posted .
Backend-heavy full-stack work on Cowork spans shared APIs, real-time agent streaming, sandboxed execution and client features across web, mobile and desktop. The role includes production ownership, observability and on-call.
Applying opens the employer's page. Want jobs like this sent to you each morning? Open Elip.
| Location | Bengaluru, Karnataka |
|---|---|
| Work mode | On-site |
| Experience | 3–4 years |
| Pay | Not stated by the employer |
| Role | Full Stack Engineer |
| Skills asked for | Python, PostgreSQL, REST APIs, API design, SQLAlchemy, Alembic, OpenAPI, Go |
| Posted | 17 Nov 2026 |
Elip's summary
Builds backend-heavy full-stack systems for a technology product team. Designs APIs, data models, real-time streaming and agent harness integrations, then ships features across web, mobile and desktop using Python, PostgreSQL and React. Operates production services with Kubernetes and observability.
The job, in the employer's words
About the role
We're hiring two backend-heavy full-stack engineers (SDE 2 or SDE 3) for the Cowork team at Sarvam. Cowork is the agentic side of Indus: users hand it real work, and an agent plans, uses tools and connectors, works in a cloud sandbox, and hands back files, docs and artifacts across web, Android, iOS, macOS and Windows.
The role is full stack, but the centre of gravity is the backend: the BFF that every client talks to, and increasingly the agent harness behind it. You'll ship features end to end, from the API and data model through to the client screens on mobile, desktop and web.
What you'll work on
Most of your time goes into the BFF and the services around it; the rest into client work and the harness, depending on what the roadmap needs that quarter.
- BFF (Chat Service): design and own the APIs that web, mobile and desktop share: sessions, conversations, streaming turns, artifacts, connectors, custom and managed agents. Data modelling, migrations, auth and RBAC, pagination and search.
- Real-time streaming: stream LLM output, tool calls and agent events from the harness to clients over an event bus, and keep it correct under reconnects, retries and partial failures.
- Agent harness and platform: contribute to Panja, our agent harness: tool and MCP integrations, sandboxed execution, file and artifact flows, latency and token-cost work.
- Clients: build features end to end on our Kotlin/Compose Multiplatform apps (Android, iOS, macOS, Windows) and the Next.js web app, so a feature you design in the backend reaches users on every surface.
- Production ownership: observability, on-call, RCAs, and making the system faster and cheaper with numbers to back it.
Tech stack
Python and PostgreSQL are the core; you don't need every item below on day one.
- Backend: Python, SQLAlchemy + Alembic, PostgreSQL, OpenAPI
- Supporting services: Go
- Mobile and desktop: Kotlin Multiplatform + Compose Multiplatform (Android, iOS, macOS, Windows)
- Web: Next.js, React, TypeScript
What we're looking for
A strong backend engineer who has also shipped real frontend work and is happy to keep doing both.
- 3+ years (SDE 2) or 5+ years (SDE 3) building and running production backend services.
- Deep in at least one backend language; Python preferred, Go, Kotlin/JVM or TypeScript/Node also welcome.
- Solid API and data-model design: REST/streaming APIs, PostgreSQL schema design, migrations, indexing, transactions.
- Distributed-systems fundamentals: async processing, queues or event streams, idempotency, retries, failure handling.
- Production experience on Kubernetes or a similar platform, with observability you actually used to debug an incident.
- Has shipped user-facing features on at least one client platform: Android, iOS, desktop or web (React/Next.js). Kotlin/Compose experience is a plus, not a gate.
- Willing to pick up the side of the stack they know less, and to learn agent and LLM systems on the job.
- Clear written communication: design docs, PR descriptions, RCAs.
Nice to have
- Built on LLM APIs: tool calling, streaming, prompt caching, evals, or an agent loop of your own.
- MCP servers or clients, OAuth connector integrations, or sandboxed code execution.
- Kotlin Multiplatform / Compose Multiplatform, or native Android/iOS.
- Real-time systems: WebSockets/SSE, NATS, Kafka, or Temporal-style durable workflows.
- Go services, or performance and cost work with measurable before/after numbers.
- Open-source work or public writing we can read.
About sarvam
- People in Elip's index who work here
- 349
- Typical time at sarvam
- 13 months
- Typical career length so far
- 4.9 years
Largest teams
- Engineering108 people
- Research16 people
- Consulting15 people
- Operations15 people
Totals from public professional profiles in Elip's index; any group under ten people is left out.
Similar open jobs
About this page
Elip lists jobs that employers publish on their own careers sites, never copies from job boards. This one comes from sarvam's listing on Ashby, posted 17 Nov 2026. The summary is Elip's; the description is the employer's. When the job closes, this page says so.