Dialpad
Senior AI Engineer (Speech) Dec 2019 - Sep 2026
- DRI for the cascaded voice agent architecture, working across multiple engineering teams, which became the default agent architecture at Dialpad
- Moved the voice agent off a realtime-only pipeline onto a cascaded design, improving accuracy and latency
- Built the latency instrumentation used to evaluate that architecture, instrumenting the incumbent system first
- Built POCs for the next in-house speech recognition service, designed for agent workloads rather than transcript readers
- Owned customer issue triage for the agent platform, then automated it with an agent built on the Claude Agent SDK
- Architected and led a next-gen Speech Recognition product from R&D to production: toolkit agnostic, streaming end-to-end ASR for conversational, telephony and videoconferencing speech under low latency and multi-accent conditions, outperforming Kaldi
- Served ASR models on NVIDIA Triton with concurrent streaming decode on a single GPU
- Benchmarked toolkits (Kaldi, K2, ESPnet, NeMo, WeNet) and end-to-end architectures (CTC, AED, Transducer, Transformer, Conformer) against hybrid models and external ASR services to choose the stack
- Developed interfaces for the shallow fusion of multi-level (sub-word and word) RNNLMs and n-gram LMs
- Developed methods to bias the models towards a list of keywords, resulting in an absolute WERR of 7%
- Post-training quantization of ASR models to achieve 50% faster RTF and 75% smaller models on disk
- Implemented the ASR inference in ONNX runtime, reducing the latency by 3x
- Mentored engineers across the speech and AI teams, ran internal talks and upskilling sessions, and helped grow the team’s depth in end-to-end ASR
Observe AI
Machine Learning Intern - ASR May 2019 - Aug 2019
- Implemented keyword-spotting models from the literature (Deep-KWS, CTC KWS) and a feature extraction pipeline in
tf.signalandtf.data - Deployed the KWS model on TensorFlow Serving with an RTF of 0.05 on GPU
IIIT-Bangalore
Research Scholar Jan 2017 - Dec 2019
- Developed end-to-end methods for multilingual and code-switching scenarios in Indian Languages
- Developed joint ASR and KWS systems using joint phoneme-grapheme recognition
- Developed a more accurate and faster training method by jointly training alignment and ASR model
- Mentored MTech and iMTech students in their projects and thesis work and delivered various tutorials and talks around ASR
Sonus Networks
SVT Engineer Aug 2015 - Jan 2017
- Sustaining engineering on real-time communication products, Sonus Insight (EMS) and SBC
- Built automated test frameworks in Python, Perl and Java, cutting team effort from hours to minutes