Senior Staff Engineer- ASR

Gnani Innovations Private Limited.; Bengaluru; On-site; INR 4,000,000 - 7,000,000; Full time

Job summary

Gnani Innovations Private Limited is seeking a Senior Staff Engineer for ASR to lead end-to-end research and production-grade speech recognition across Indian languages, focusing on accuracy, multilingual coverage, and low-latency streaming.

You’ll own architecture choices, mentor engineers, publish findings at conferences, and translate research into scalable enterprise systems for real-world voice applications.

Qualifications

• 8–10 years in speech / ML research or engineering, with substantial contribution to production-grade ASR systems.

Responsibilities

• Advance ASR research for Indian languages -- accuracy, multilingual coverage, and robustness across modern architectures.
• Ship streaming and offline ASR with low latency, high throughput, and reliability at scale.
• Handle real-world audio and build the real-time components conversations depend on.
• Technical leadership – Guide engineers, design rigorous experiments, and own the standard for how ASR gets built here.
• Publish and represent Gnani’s ASR work externally -- international conferences, talks, and community engagement.

Skills

Automatic Speech Recognition

ASR Pipeline Workflows

Voice AI

Advanced Speech Architectures & LLMs

Real-time ASR architecting

Tools

PyTorch

NVIDIA NeMo

ESPnet

Hugging Face

Kaldi / k2

Job description

Senior Staff Engineer, Automatic Speech Recognition (ASR)

About the role

While you’re reading this, Gnani is listening to thousands of customers across India — in Hindi, Tamil, Telugu, and dozens more languages, over noisy phone lines, thick accents, and mid-sentence code-switches. That understanding is our ASR stack. It’s the difference between a voice agent that mishears and one that gets it right the first time, in the language and accent the caller actually uses.

We’re building the most accurate speech recognition for Indian languages, in real time and at enterprise scale — and we want you to lead the research behind it. As Senior Staff Engineer for ASR, you own the stack end-to-end: advancing accuracy and multilingual coverage, engineering it for streaming latency, and hardening it for the messy realities of production audio. This is a senior individual-contributor role: you set technical direction, design experiments, and mentor the engineers around you — without a people-management load.

Core mandate

• Research – Advance ASR research for Indian languages -- accuracy, multilingual coverage, and robustness across modern architectures.
• Production & performance – Ship streaming and offline ASR with low latency, high throughput, and reliability at scale.
• Handle real-world audio and build the real-time components conversations depend on.
• Technical leadership – Guide engineers, design rigorous experiments, and own the standard for how ASR gets built here.

What you’ll drive

Research & modeling

• Own the ASR modeling roadmap -- architecture selection, accuracy, and multilingual/code-switched coverage across Indian languages.
• Work across modern ASR architectures (Conformer, E-Branchformer, RNN-T, CTC, Whisper/encoder-decoder, and Speech Language Models) and translate findings into shippable systems.
• Push on the hard multilingual problems: code-switching, regional accents, dialects, pronunciation variation, and low-resource languages.
• Train and fine-tune large-scale multilingual models -- data collection, augmentation, domain adaptation, contextual biasing, and custom vocabulary.
• Own the text side of recognition: tokenization (BPE, SentencePiece, phonemes), language modeling, confidence scoring, punctuation, capitalization, and text normalization.
• Participate in benchmarks and publish the work in international conferences.

Production & performance engineering

• Design ASR systems for production -- streaming (low partial / first-token latency) and offline (high-throughput batch), with favourable RTF.
• Optimize production inference using ONNX, TensorRT, NVIDIA Triton, quantization, batching, and GPU-efficient serving architectures.
• Partner with platform and infra teams on serving, scaling, and reliability -- latency SLAs, uptime, error budgets.
• Build evaluation and regression harnesses so accuracy and latency don’t silently regress release to release.

Robustness & real-time Voice AI

• Build robust ASR for real-world environments -- noisy audio, telephony, far-field speech, reverberation, and overlapping speakers.
• Develop the Voice AI components real-time conversations depend on: VAD, endpointing, turn detection, barge-in, speech segmentation, and conversational state management.
• Turn recurring field failures (accent, domain, channel) into systematic model and pipeline improvements.

Technical leadership & research presence

• Technically guide engineers -- experiment design, tracking, code and research review, and career-shaping mentorship.
• Translate ASR research into enterprise-scale production systems and define the architecture that gets there.
• Set the bar for rigor: reproducible experiments, clean baselines, honest evaluation, and clear write-ups.
• Stay current with ASR research and bring the relevant frontier into Gnani’s roadmap.
• Publish and represent Gnani’s ASR work externally -- international conferences, talks, and community engagement.

Who you are

Experience we’re looking for

• 8–10 years in speech / ML research or engineering, with substantial contribution to production-grade ASR systems.
• A track record of designing, training, and deploying streaming and offline ASR at scale -- high accuracy, low latency, high reliability.
• Deep expertise in multilingual ASR, especially Indian languages -- code-switching, regional accents, dialects, pronunciation variation, and low-resource languages.
• Strong command of modern ASR architectures -- Conformer, E-Branchformer, RNN-T, CTC, Whisper/encoder-decoder, and Speech Language Models.
• Good publications at international conferences (Interspeech, ICASSP, NeurIPS, ICML, ACL) in speech / audio / ML.
• A proven record of translating ASR research into enterprise-scale production systems and mentoring high-performing engineering teams.

What makes a standout candidate

• Experience building ASR for hostile audio -- telephony, far-field, reverberation, and overlapping speakers.
• Hands‑on with real-time Voice AI components -- VAD, endpointing, turn detection, and barge-in.
• Depth in tokenization, language modeling, contextual biasing, and text normalization.
• Familiarity with the Indian voice landscape and low-resource data strategies.
• Comfortable operating with high ownership in a fast-moving, post–Series B environment.

Technical fluency -- a must-have

• Hands‑on with PyTorch, NVIDIA NeMo, ESPnet, Hugging Face, and Kaldi / k2, plus distributed training and large-scale evaluation.
• Fluency with ASR evaluation -- WER (and CER), latency, RTF -- and what drives improvements or regressions on each.
• Production inference optimization -- ONNX, TensorRT, NVIDIA Triton, quantization, batching, and GPU-efficient serving.
• Judgment on evaluation and benchmark design: good test sets, fair comparison across languages / accents / domains / channels, and not overfitting to a benchmark.

Why Gnani, why now

Gnani.ai is one of the few companies in India doing original speech research at depth -- our models are trained on real Indian language data at scale, not adapted from English-first architectures. As Senior Staff Engineer for ASR, you’ll own one of the most technically demanding surfaces in Indian voice AI, with the freedom to set direction and the impact of seeing your models understand real conversations across banking, healthcare, logistics, and government services.

High visibility, high impact, and a clear path to the most senior technical roles as the company scales.

Skills required

• Automatic Speech Recognition (ASR)
• ASR Pipeline Workflows
• Voice AI
• Advanced Speech Architectures & LLMs
• ASR real-time speech architecting