Piyush Vyas

resume

download pdf

Experience

Deep Learning Architect / Delivery Lead

Amazon Web Services

May 2024 to Present

  • Led 30+ GenAI production deployments for Fortune 500 and startups, averaging 8-week time-to-value.
  • Architected and coded agentic AI, RAG, IDP, conversational AI, and browser automation systems.
  • Developed 15+ reference architectures and production code samples adopted network-wide.
  • Delivered 35+ executive workshops for C-suite and engineering leaders.
  • Orchestrated cross-functional teams of 4 to 6 including consultants, forward-deployed engineers, and applied scientists.
  • Drove 40 to 50% operational cost savings for customers.
  • Accelerated partner GTM through reusable assets and enablement programs.

Senior Machine Learning Engineer

Numerade Labs

Apr 2023 to May 2024

  • Owned the AI product roadmap and technical strategy across all ML initiatives.
  • Built an AI video generation system from scratch (image gen, animation, TTS, avatars).
  • Deployed LLMs via vLLM achieving 10x throughput and 60% memory reduction.
  • Built and managed a team of 4 ML engineers and 2 interns; promoted 1 to senior within 12 months.
  • Led B2B adoption across 50+ academic partners achieving 90%+ feature adoption.
  • Delivered 70% cost reduction, 30% adoption increase, and 45% engagement lift.

Senior Machine Learning Scientist

Spectrum Labs

Oct 2021 to Apr 2023

  • Built proprietary LLMs from scratch including fine-tuning, evaluation, quantization, and deployment.
  • Created a PyTorch toolkit adopted by the entire data science team.
  • Led a team of data scientists and ML engineers; established research-to-production processes.
  • Scaled the system to 12+ languages processing billions of daily messages.
  • Achieved a 40% latency reduction at 98% accuracy.
  • Partnered with the CTO and GTM on product strategy influencing multi-million-dollar ARR.

Research Engineer

Comcast Applied AI

Aug 2021 to Oct 2021

  • Built a production ASR system using wav2vec 2.0 achieving 30% WER reduction.
  • Served millions of users.
  • Published at IEEE ICASSP 2022.

Research Engineer

Indiana University Bloomington

May 2020 to Aug 2021

  • Led research in speech translation and assessment.
  • Published at INTERSPEECH and IEEE ICASSP, winning 2 best-paper awards.

Education

  • M.S. in Intelligent Systems Engineering Indiana University Bloomington
    2020
  • Post Graduate Diploma, Computer Applications Devi Ahilya Vishwavidyalaya
    2018
  • Bachelor of Commerce Devi Ahilya Vishwavidyalaya
    2017

Publications

  • Optimally Encoding Inductive Biases into the Transformer Improves End-to-End Speech Translation

    Piyush Vyas, Anastasia Kuznetsova, Donald S. Williamson

    INTERSPEECH 2021 · Best Student Paper Award · Read PDF

  • An End-to-End Non-Intrusive Model for Subjective and Objective Real-World Speech Assessment Using a Multi-Task Framework

    Zhuohuang Zhang, Piyush Vyas, Xuan Dong, Donald S. Williamson

    IEEE ICASSP 2021 · Outstanding Student Paper Award · Read PDF

  • Temporal Early Exiting for Streaming Speech Commands Recognition

    Raphael Tang, Karun Kumar, Ji Xin, Piyush Vyas, Wenyan Li, Gefei Yang, Yajie Mao, Craig Murray, Jimmy Lin

    IEEE ICASSP 2022 · Read PDF

Open source

Ten public projects across the agent reliability layer (Browser Agent Protocol, DBAR, uSEID, skill-tools, agent-contracts, agent-pager) and earlier speech AI research (SpecAugment with 94 stars, GraphML, VoiceID, WhisperingGPT). Full list on projects.

Skills

Engineering
Python, TypeScript, JavaScript, Swift, PyTorch, FastAPI, Next.js, Astro, Playwright, Zod, OpenTelemetry, Docker, BM25
AI / ML
vLLM, LangChain, Hugging Face Transformers, LLM fine-tuning and evaluation, quantization, on-device WASM, model deployment, distributed training
Cloud & Platforms
AWS (Bedrock, SageMaker, Lambda, CDK), OpenAI API, Anthropic API, Google Gemini, Vector Databases, Open Food Facts
Agent & Browser
Agentic AI, multi-agent orchestration, MCP servers, browser automation, accessibility-tree selectors, deterministic replay (CDP virtual time, network record/replay), agent contracts, RAG, IDP, conversational AI
Speech & Audio
wav2vec 2.0, Whisper, ASR systems at scale, speaker verification, speech translation, speech assessment, speech synthesis, multi-task learning
Product & Solutions
GenAI solution architecture, reference architecture design, production deployment, cost optimization (40 to 70%), eight-week time-to-value delivery, customer discovery, roadmap ownership, B2B adoption strategy, executive presentations
Delivery & Leadership
Cross-functional team orchestration (4 to 6), engineering management, hiring and promotion, mentorship, 35+ executive workshops delivered, C-suite engagement, partner GTM, research-to-production process design, multi-million-dollar ARR strategy