Backend Engineer - Vision

Role Overview

Sarvam is building the bedrock of Sovereign AI for India, developing a full-stack platform across research, models, infrastructure, and applications. The Backend Engineer - Vision will join the research teams to build the serving harness that transforms in-house vision-language models into a production-grade document intelligence platform. You will build and own services that sit between vision models and enterprise consumers, managing complex pipelines involving ingestion, inference, and validation.

Responsibilities

• Build and maintain REST APIs for document submission, job status, and result retrieval for both synchronous and long-running async flows.
• Implement Temporal workflows and activities for multi-stage document pipelines including splitting, pre-processing, inference, and assembly.
• Write pre- and post-processing logic for page segmentation, de-skewing, layout handling, and schema validation.
• Instrument systems for latency and cost observability using OpenTelemetry-based approaches.
• Build internal tooling such as replay harnesses, evaluation runners, and regression suites.
• Debug production issues including stuck workflows, GPU queue backpressure, and malformed inputs.
• Collaborate with the models team to translate model behavior into harness behavior.

Requirements

• 1-2 years of experience building backend services that have run in production.
• Strong fundamentals in Go or Python with the ability to write clean, concurrent, and tested code.
• Solid grasp of HTTP and REST API design, and async/background job processing.
• Comfort with PostgreSQL and Redis, including transactions and indexing.
• Working familiarity with Docker and exposure to Kubernetes.
• A debugging mindset focused on logs, traces, and code.
• Genuine interest in AI systems engineering and productionizing models.

Skills

• Go
• Python
• Temporal
• PostgreSQL
• Kubernetes

Nice to Have

• Exposure to Temporal or other durable workflow engines like Airflow, Cadence, or Step Functions.
• Hands-on work with OCR, Computer Vision, or LLM/VLM inference.
• Familiarity with GPU serving stacks such as vLLM, TensorRT-LLM, Triton, or SGLang.
• Open-source contributions or relevant side projects.