Building reliable AI systems from research to production
I am a Principal AI Scientist with nearly two decades in ML/AI, focused on speech, language models, retrieval systems, and agentic architectures at production scale. My new book, Building Speech AI, is now available.
New book
Building Speech AI
A practitioner's guide to speech recognition, synthesis, audio language models, real-time voice agents, and production tradeoffs, with runnable Python companion code.
Highlights
A quick snapshot of current work, output, and research service.
About Me
Deepak has built and shipped large-scale machine learning systems across retail, FMCG, search relevance, computational advertising, and voice assistants. He has led work on speech recognition and language understanding at Amazon Alexa, helped build contextual advertising systems at Amazon Ads, and is the author of Building Speech AI.
His work at Alexa included personalized, self-healing error-correction models that learn from user interaction data, enabling runtime query rewriting with low-latency inference paths. His technical focus connects research and production deployment across LLMs, multimodal systems, retrieval-augmented generation, and agents.
He is a Senior Member of IEEE (Seattle Chapter), reviewer for NeurIPS, ICML, KDD, AMLC, ARA, and EMNLP, and advisor to early-stage startups in advertising and healthcare.
- Microsoft
- Amazon
- Alexa
- Amazon Ads
- Freshworks
- Snapdeal
- InMobi
- Mu Sigma
- Infosys
Past life: Management consultant for a Fortune-50 CPG/pharma company, advising one of the largest loyalty program rollouts.
Research Interests
- Speech (ASR)
- Language Models
- NLP/G
- LLM Agents and World Models
- Behavioral Science
- Computational Advertising
- Search Relevance
- Information Retrieval/RAG
- Voice Assistants