Keep persevering.
Hi 👋 I'm Ayush. A builder at Amazon, on the team responsible for Alexa's personality and emotional intelligence.
Manifest.
Contents
- IProfilep. 003
- IIExperiencep. 017
- IIIWritingp. 042
- IVCorrespondencep. 071
Profile.
“Building personality and emotional intelligence for a voice assistant.”
Engineer, based in Seattle since 2022 — by way of Bloomington (for the degree) and Bombay (for everything before that). At Amazon, on the team responsible for Alexa's personality and emotional intelligence.
The work: prompt evaluation at volume, retrieval for traits the model must never contradict, audio signals for sarcasm, and the unglamorous audit of why a model said the wrong thing on a Tuesday.
Outside of that — an interest in how AI-assisted tooling changes what a single engineer can build in a weekend.
Experience.
Reverse chronological; trimmed for readability.
The Ledger · Positions Held
- DatesAug 2022 — presentPlaceSeattle, WAAug 2022 — present · Seattle, WA
Software Development Engineer II
Alexa Personality, Amazon.com Inc.
- 01I work on the language model that gives Alexa+ its personality — shaping how it speaks, what it refuses to say, and how it handles the awkward moments in between.
- 02Built the evaluation harness we use to test the model at scale, plus a retrieval layer that keeps Alexa consistent about who she is across sessions.
- 03Spend a fair amount of time in the weeds of prompt engineering, audio sentiment signals (sarcasm, frustration), and the unglamorous audit of why a model said the wrong thing to someone on a Tuesday.
- DatesMay 2021 — Jul 2022PlaceBloomington, INMay 2021 — Jul 2022 · Bloomington, IN
Research Software Engineer
Institute for the Study of Contemporary Antisemitism, Indiana University
- 01Created the backend for a low-code annotation portal used to label antisemitic data for the ISCA (Institute for the Study of Contemporary Antisemitism). Django, multi-user workflows, and a lot of careful taxonomy work.
Selected Projects · Filed on GitHub
RAGMagic
Bring your own data. Get LLM-ready context in three lines of code.
A small library that turns arbitrary documents into retrieval-ready chunks, embeds them, and exposes a minimal query API — the boring plumbing every RAG project rewrites from scratch.
web-scraper-for-ai-agents
Scrape, clean, and enrich the web for LLM consumption.
A pipeline that fetches pages, strips the noise, and emits structured, citation-ready documents — built because every agent I wrote kept choking on raw HTML.
voiceBasedGrocerySystem
A voice-driven grocery list, for the smallest kitchen.
A weekend experiment in voice UX — speech to list, list to cart, without the cognitive overhead of a screen. A prototype for how conversation replaces tapping.
Writing.
Click the entry below to read it in place; the full archive lives at /writing.
An opening dispatch.
The opening of what is intended to be a regular column — on building voice assistants, the small embarrassments of prompt engineering, LLM evaluation in practice, and whatever else seems worth writing down.
Particulars
- Filed
- 16 April 2026
- Topics
- AI · Engineering · The small things
- Reading
- A few minutes
Subsequent entries, as they appear.
Correspondence.
The addresses at which the author may reasonably be reached.
Addresses