Sarvam is building the bedrock of Sovereign AI for India. The company is developing India’s full-stack sovereign AI platform, building across research, models, infrastructure and applications with a singular focus on making AI genuinely work for India. Sarvam works with leading enterprises and public institutions and is backed by Lightspeed, Peak XV, and Khosla Ventures. Sarvam partners with India’s leading brands, including Tata Capital, SBI Life, CRED, IDFC, and LIC.
About the Role
You’ll work on the data and ML infrastructure that powers Sarvam’s dubbing pipeline — building systems that acquire, process, and annotate speech data at scale. Depending on your interest and strengths, your work will span one or more of:
Audio/Video acquisition — Automating content download, format handling, and audio extraction with robust failsafe logic
Speaker diarization — Running and validating models that determine “who spoke when” in multi-speaker audio
Emotion & style tagging — Classifying speech segments by emotion and speaking style to produce training data for expressive TTS models
This is hands-on pipeline engineering — you’ll write Python scripts that call ML models, process audio files, handle messy real-world data, and produce clean, structured outputs that downstream models consume.
What You’ll Do
Build and maintain audio/video processing scripts using tools like FFmpeg, yt-dlp, and Python audio libraries
Run ML model inference (diarization, emotion recognition) on audio data and validate outputs
Handle edge cases: failed downloads, format mismatches, model errors, rate limits, and scale constraints
Process and structure outputs (RTTM files, JSON annotations, CSVs) for use in model training
Think about reliability, idempotency, and logging so pipelines can run unattended on thousands of files
What We're Looking For
Comfort with Python scripting — loops, file I/O, subprocess calls, basic error handling
Willingness to learn audio concepts: sample rates, WAV/MP3 formats, mono/stereo
Ability to read documentation and get unfamiliar libraries/models working
Problem-solving instinct — when something breaks, you debug rather than give up
Familiarity with using AI coding tools (ChatGPT, Copilot, Claude) to accelerate your work
Bonus Points
Experience with any audio/ML library (librosa, torchaudio, soundfile, HuggingFace transformers)
Familiarity with FFmpeg or command-line media tools
Exposure to ML model inference (loading a model, running predictions)
Basic understanding of speech/NLP concepts (ASR, TTS, diarization)