Sashank Kommareddy

About Sashank Kommareddy

I have over ten years of experience executing technology programs in Retail, Telecom, and Advertising businesses. I am currently a Senior Technical Program Manager in Alexa AI core platform team that comprises of AI research scientists and engineers. My focus is on developing and deploying advanced multimodal large language models & AI systems that power the conversational AI assistant - Alexa+, and further enhance Agentic voice based commerce capabilities. Prior to this, i\'ve led highly complex cross functional programs to convert research ideas into commercially scalable AWS physical retail products for autonomous checkout and biometric payments using technologies such as deep learning, computer vision, and sensor fusion.I am passionate about transforming cutting-edge technologies into scalable solutions that enhance customer experiences and generate economic value.

Experience

1. Sr Technical Program Manager - Conversational Ai Modeling & Learning

   Amazon

   Apr 2024 — Present · Austin, TX, US

   Alexa+ core AI model and infrastructure platform that operates a proprietary conversational AI stack spanning foundational models, LLM serving infrastructure, and an integrated training-evaluation platform supporting 80+ feature teams across Amazon Music, Shopping, Prime Video etc- Delivered 45% net runtime GPU efficiency improvement (~$150 million annual operations cost reduction) by executing complex cross org programs with Alexa, AWS, AGI, and Nvidia to launch custom built vLLM and upgrades on existing TRT containers for various models used by Alexa- Accelerated productionization of first 3P open weight model GPT-OSS 120B in weeks by developing a model evaluation, and prompt optimization framework along with automated developer tooling for both science and engineering teams- Managed a cost performant proprietary model development program involving science and engineering teams from Alexa and AGI orgs using novel telescopic zoom attention architecture, custom training tools to implement Reinforcement Learning with Verifiable Rewards (RLVR) and TRT containers with Prompt Caching, Prefill Chunking, and Speculative Decoding with EAGLE3. This led to 40% optimization in both Latency as well as Inference throughput efficiency on Nvidia H100 and H200 GPUs- Transformed program management operations through AI-enabled execution model and an organizational intelligence layer encoded in Git-versioned instruction files, reducing weekly operational overhead from 15-20 hours to 3-4 hours (75-80% time savings). Developed 25+ version-controlled AI instruction files that automate executive reporting, quality assessment, and cross-program analysis across 20 strategic initiatives spanning three pillars: Alexa+ Optimization, Next-Gen Architecture, and Model Innovation

Education

• SMU Cox School of Business

  Master of Business Administration - MBA, Finance and Entrepreneurship

• Jawaharlal Nehru Technological University

  Bachelor of Technology - BTech, Computer Science

Skills

• Agile Methodologies
• Management
• Software as a Service (Saas)
• Oracle
• Business Analysis
• Telecommunications
• Analysis
• Sql
• Web Technologies
• Business Intelligence
• Product Strategy
• Cross-Functional Team Leadership
• Erp
• Product Management
• Corporate Finance
• Data Analysis
• Project Management
• Strategy
• Leadership