AI Engineer

You will own AI systems end-to-end, from the speech-to-text models that turn audio into text, to the diarization layer that separates speakers, to the agentic layer that converts conversation into memory and action, to the observability and evaluation infrastructure that keeps all of it honest in production. This is a wide role by design. You own model selection, serving, and production reliability. If you want to fine-tune one model and ignore the system around it, this is not the role.

The candidate will have responsibilities across the following functions:

What You Will OwnSpeech-to-Text:

• Evaluate, integrate, and optimise STT models across cloud and self-hosted options.
• Drive accuracy and cost trade-offs grounded in ground-truth WER metrics.
• Own Whisper fine-tuning, Indic normalisation, and domain-specific adaptation.

Speaker Diarization and Identification:

• Push DER accuracy on hard, real-world multi-speaker audio.
• Own diarization pipeline from pyannote integration to production serving.
• Build labelled benchmark datasets reflecting real conversational audio.

Agentic AI:

• Build the memory and retrieval pipeline that sits on top of captured conversation.
• Design and own LLM orchestration using LangGraph or equivalent frameworks.
• Ship agent workflows that convert transcript output into structured memory and action.

Model Serving and Infrastructure:

• Stand up and optimise self-hosted model serving: vLLM, Triton, or equivalent.
• Own latency, throughput, and cost-per-user targets in production.
• No API-only serving; you run the infra.

Observability:

• Instrument the full audio-to-memory pipeline: STT, diarization, retrieval, and LLM calls.
• Define and track model-quality SLOs: transcription drift, DER over time, retrieval relevance, latency.
• Build dashboards and alerting so model degradation is caught before users feel it.
• Trace failures across a distributed, always-on system using metrics, logs, and traces.
• Close the loop: production signals feed back into evaluation and model selection.

Evaluation:

• Build and own ground-truth eval harnesses for every model in the stack.
• Measure with real metrics: WER for transcription, DER for diarization, Recall and F1 for retrieval.
• Run regression and A/B evaluations on every model swap, prompt change, or pipeline update.
• Nothing ships on a vibe; reject anecdotal proxies and cherry-picked examples as evidence of quality.

Requirements:

• 3 to 5 years as an AI/ML engineer with production systems behind you.
• Agentic AI in production: LangGraph, multi-agent pipelines, tool-calling, memory systems.
• Self-hosted model serving vLLM, Triton, or equivalent; not just API calls.
• LLM gateway and orchestration patterns you have designed and shipped.
• Strong software engineering: you write code that survives contact with real users.
• Fluency in Python and the production ML ecosystem.
• Observability and evaluation discipline: you measure first and trust metrics over intuition.
• Tier-1 college background: IIT / NIT / BITS / IISc.

Strong Signal (Not Required):

• Extracurricular academic spike: Olympiad rankings, ICPC, KVPY, Kaggle Expert/Master.
• Audio and speech ML experience: STT, diarization, voice pipelines.
• Research background or publications in ML, NLP, or speech.
• Experience building eval harnesses or production model-monitoring systems.
• GCP and Kubernetes comfort.