Sarvam
About this role
Sarvam is building India's sovereign AI platform, backed by Lightspeed, Peak XV and Khosla Ventures, with customers including Tata Capital, SBI Life, CRED and LIC. Studio is its media platform covering AI dubbing, live translation, voice cloning, stem separation, lip sync and music generation. This backend role spans three codebases: a core ML pipeline library handling speech recognition, translation, text to speech and audio processing, the production services for dubbing and live translation, and a shared platform SDK used by every Studio service. You design FastAPI services with multi stage task orchestration and backpressure control, build distributed worker architectures that scale per pipeline stage and recover stuck tasks, own the async ORM data layer on PostgreSQL, and implement WebSocket job tracking and streaming audio. Python and async programming depth is non negotiable here, and media processing experience with FFmpeg is expected.
Who this is for
What the posting asks for: 4 to 6 years in backend engineering with a focus on building and operating production services at scale. Sarvam marks Python and production FastAPI or similar async web service experience as non negotiable. Beyond that it asks for a deep understanding of async programming including asyncio, concurrent execution patterns and designing for high throughput, experience with distributed task systems covering task queues such as Celery, message brokers and fault tolerant job orchestration, hands on PostgreSQL with an async ORM such as SQLAlchemy including query optimisation, schema design and migrations, familiarity with audio and media processing using FFmpeg, common audio formats and libraries such as soundfile or librosa, experience integrating ML models into production whether API based over REST or gRPC or local inference with PyTorch or ONNX Runtime, experience building reusable libraries or SDKs with clean APIs and backward compatibility, proficiency with Docker, Kubernetes, Helm and at least one major cloud, and strong testing discipline including mocking external services and maintaining CI/CD.
Bonus points: prior experience with speech or NLP systems such as ASR, TTS or machine translation.
The real day to day, across three areas. Service and infrastructure: designing production FastAPI services for dubbing and live translation with multi stage task orchestration, rate limited scheduling and backpressure controls, building distributed worker architectures with independent scaling per pipeline stage and automatic recovery of stuck tasks, owning the async ORM data layer and migrations on PostgreSQL, implementing WebSocket based job tracking and streaming audio pipelines, and managing Kubernetes deployments including Helm charts, secrets, ingress and multi role images. ML pipeline and library: extending the core dubbing library across audio extraction, voice activity detection, speech recognition, translation, quality control, text to speech and video stitching, integrating and optimising ML model serving for both remote inference and local inference, building quality control orchestration including automated scoring, tempo analysis and pronunciation verification, and maintaining LLM integration layers across multiple provider backends. Platform SDK: maintaining the shared Studio SDK with reusable FastAPI middleware and routers for authentication, billing, workspace isolation and input validation.
Location and setup: Bengaluru. The posting does not state office days.
Honest fit guidance: the media and audio processing element is a genuine specialisation, and the posting is explicit that FastAPI and async depth are not negotiable. If your backend experience is synchronous request and response web services, this will be a real jump. Sarvam also has a Studio DevOps role open today at 3 or more years, which is the same platform from the infrastructure side.
Bonus points: prior experience with speech or NLP systems such as ASR, TTS or machine translation.
The real day to day, across three areas. Service and infrastructure: designing production FastAPI services for dubbing and live translation with multi stage task orchestration, rate limited scheduling and backpressure controls, building distributed worker architectures with independent scaling per pipeline stage and automatic recovery of stuck tasks, owning the async ORM data layer and migrations on PostgreSQL, implementing WebSocket based job tracking and streaming audio pipelines, and managing Kubernetes deployments including Helm charts, secrets, ingress and multi role images. ML pipeline and library: extending the core dubbing library across audio extraction, voice activity detection, speech recognition, translation, quality control, text to speech and video stitching, integrating and optimising ML model serving for both remote inference and local inference, building quality control orchestration including automated scoring, tempo analysis and pronunciation verification, and maintaining LLM integration layers across multiple provider backends. Platform SDK: maintaining the shared Studio SDK with reusable FastAPI middleware and routers for authentication, billing, workspace isolation and input validation.
Location and setup: Bengaluru. The posting does not state office days.
Honest fit guidance: the media and audio processing element is a genuine specialisation, and the posting is explicit that FastAPI and async depth are not negotiable. If your backend experience is synchronous request and response web services, this will be a real jump. Sarvam also has a Studio DevOps role open today at 3 or more years, which is the same platform from the infrastructure side.
Apply on company site
Opens jobs.ashbyhq.com, the employer's own application page. Applying is always free.