Available for opportunities

Hi, I'm Ayush Sanghavi

Software Development Engineer at Amazon building Alexa's personality and emotional intelligence with LLMs, RAG systems, and prompt engineering.

scroll

Building intelligent
systems at scale

I'm a Software Development Engineer II at Amazon, where I work on Alexa's personality and emotional intelligence. I design and build systems that make AI assistants more human — from emotional understanding through audio sentiment analysis to RAG-powered personality retrieval.

My work spans the full stack of modern AI engineering: LLM evaluation platforms, retrieval-augmented generation, prompt engineering, and building production systems that serve millions of users daily.

I hold a Master's in Computer Science from Indiana University Bloomington and am an AWS Certified Cloud Practitioner. I'm passionate about leveraging AI-assisted development tools to push the boundaries of what's possible.

4+
Years of Experience
1M+
Prompts Evaluated
5+
Services Migrated
90%
Defect Reduction

Where I've worked

AUG 2022 — PRESENT
Software Development Engineer II
Alexa Personality · Amazon.com Inc · Seattle, WA
  • Designed and deployed "PersonaLab," a serverless LLM evaluation platform that processed over 1M prompt utterances to benchmark Alexa+ response quality, slashing manual verification time by 200+ hours.
  • Developed Alexa's emotional intelligence by integrating audio-based sentiment models with a custom signal algorithm that detects sarcasm through high-confidence frustrated audio tones paired with positive lexical text.
  • Built a RAG retrieval API with OpenSearch k-NN vector queries, ensuring 100% consistency for immutable personality traits like identity and preferences.
  • Reduced false "roast" rate by 90% (2.3% → 0.25%) through prompt engineering and comprehensive regression testing.
  • Pioneered Alexa Personality Service exposure to Mobile App for the launch of Alexa Styles, driving regulatory compliance and API security review.
  • Led migration of 5+ service infrastructure to JDK 17/19 using Amazon Q and Kiro.
MAY 2021 — JUL 2022
Software Engineer
Institute for the Study of Contemporary Antisemitism · Indiana University
  • Built a Python automation platform to extract 40K tweets via Twitter API and capture 5K screenshots for archival storage on Google Drive.
  • Developed and maintained a multi-user annotation portal using Django, Bootstrap, and HTML/CSS with user registration and email confirmation workflows.

Things I've built

🧠

PersonaLab — LLM Evaluation Platform

Serverless platform that benchmarks Alexa+ response quality across 1M+ prompts using an automated "LLM-as-a-Judge" scoring system and a low-code UI for cross-functional science teams.

AWS Lambda LLM Evaluation Serverless
🎭

Alexa Emotional Intelligence

Sarcasm detection system combining audio sentiment analysis with lexical text scoring. Triggers dynamic prompt injection for context-aware empathetic responses when sarcasm is detected.

Sentiment Analysis Prompt Engineering NLP
🔍

RAG Personality Retrieval API

Converts user utterances into dimensional vectors for OpenSearch k-NN queries, deterministically overriding parametric LLM generation for 100% consistency on immutable personality traits.

RAG OpenSearch k-NN Vector Search
🌦️

Weather Forecasting System

Microservices-based architecture displaying atmospheric charts from NEXRAD and NASA MERRA-2 satellite data. Stress-tested to handle thousands of requests per minute at 95% success rate.

Java React Docker Kubernetes Jenkins

Technologies & tools

Languages

Java Kotlin Python SQL

AI / LLM

RAG Systems Vector Search OpenSearch k-NN Prompt Engineering LLM Evaluation Agentic Workflows Function Calling MCP

AI-Assisted Development

Claude Code Cline Kiro Amazon Q

Cloud & DevOps

AWS Docker Kubernetes CI/CD Jenkins

Frameworks & Tools

React Django PostgreSQL RabbitMQ Git

Certifications

AWS Certified Cloud Practitioner

Let's connect

I'm always open to discussing new opportunities, interesting projects, or just having a great conversation about AI and engineering.