Software & AI Engineer at Amazon

Gabe
Mehra

Harvard UniversityBoston, MAAmazon SDE

I'm a software and AI engineer at Amazon, where I work on Alexa Edge AI Services. I studied Computer Science and Music at Harvard University (graduated December 2024). I love music, and I'm passionate about how technology can advance artistic expression.

Gabe Mehra
01

Experience

Software Development Engineer

March 2025 - Present
Amazon · Alexa Edge AI Services · Boston, MA
  • Owned an end-to-end testing framework for Alexa's Follow-Up-Mode enablement Lambda, driving it across dependencies to completion.
  • Automated data pipelines for Alexa's automatic media detection, saving ~10 developer-hours weekly.
  • Contributed technical judgment to 15 design-document reviews, evaluating space/time complexity tradeoffs, extensibility, and edge-case/error handling across components owned by other teams.
  • Consistently applied AWS best practices (the AWS Golden Path) across designs and implementations.

Software Engineer Intern

June - August 2024
Amazon · Alexa Edge Devices · Boston, MA
  • Reduced single-feature cost by ~86% by relieving cloud-compute strain (C++, Python).
  • Built the case, with data, that Alexa's whisper detection is best run on-device, and presented the recommendation to a multi-team panel of Amazon engineers and managers.
  • Demoed a new feature on an Alexa device to cross-functional stakeholders.

Generative AI Music Course Assistant

January - December 2024
Harvard University · Bok Center Learning Lab · Cambridge, MA
  • Designed a Transformer model and post-processing pipeline to generate MIDI scores, trained on representational music to predict the next note rather than the next word.
  • Integrated the tool with a piano keyboard for live input processing.

ML and Software Engineering Intern

May - August 2023
Beaver Health · Palo Alto, CA
  • Shipped a production LLM-based AI agent end to end for a clinical research setting (NIH-funded R&D), delivering an interactive agent with a user-facing interface at an early-stage startup.
  • Engineered the agent's safety guardrails using input/output logistic-regression and LLM classification to keep the autonomous system on-track, improving correct classification by 50-60%.
  • Shaped the product roadmap and built the full stack: Python, Flask, JavaScript, TypeScript, Node.js, PostgreSQL, Google Cloud, LLM API integrations.

Associate Software Development Intern

June - August 2022
MACH37

MACH37 is an accelerator company that specializes in very early stage cybersecurity startups. While I was working there, I built a dynamic web application for MACH37 to run their Fall and Spring 2022 accelerator programs. The web app includes a Calendly style scheduling feature between two users, a Google Forms style assignment feature, and a feature that allows other types of users to give feedback ratings on those assignments.

Insurance Intern

June - August 2021
REX

REX is a company that guides its clients, everyday citizens, through the process of buying a home. While there, I worked with the insurance team creating reports and analytics for the company to help reach insurance policy attachment goals and improve company yield and service usage. I automated several internal insurance processes within Salesforce and I improved upon current internal insurance team work-flows. I also identified problems in data storage / data cleanliness.

02

Projects

Prediction Market Making Bot

October 2026 - Present
PythonDockerAWS

An automated market-making system for prediction markets that continuously quotes both sides of a market. Built around a plug-and-play architecture where pricing, risk and exchange components can be swapped through configuration. Includes a replay engine for testing changes against recorded market data, side-by-side comparisons of strategy variants, built-in risk limits and safety stops, and a performance report that breaks down where profit and loss come from. Runs 24/7 in the cloud.

HTMLJavaScriptPolymarket APIGitHub Pages

A 16-bit style pick'em game for the 16 most competitive 2026 U.S. Senate races. Pulls live polls (with polling averages calculated in the page) and live prediction-market odds from Polymarket, and falls back to a built-in snapshot if either source is down. A single static page, live at senatepicks.com.

The Harvard Callbacks Network ↗

December 2024 - January 2025
FirebaseNextJSTypeScriptJavaScriptTailwindPrimeReact

A networking platform for alumni and undergraduate members of my college a cappella group, The Harvard Callbacks, to connect and search via location and field of work. Contains an anonymous emailing feature and personalized profile editing.

The Harvard Callbacks Network

Text to Music Machine Learning Tool

Fall 2024
PythonPyTorch

LoRA fine-tuning of a Contrastive Learning guided Stable Audio Diffusion model using the MusicCaps dataset.

Musical Melody Completion Tool ↗

December 2023 - May 2024
PythonTensorFlow

A tool for musicians to complete their musical ideas. Transformer Model that predicts the next note instead of the next word. Trained on several hundred German Folk songs to complete a melodic thought. Outputs to a Musescore musicxml file.

Python

A Vocal Training Game! Uses pitch detection software to track the user's pitch. Calculates score by awarding more points for a more accurately sung note. Sing and match the notes on screen by seeing the relative position. Can choose up to three different songs!

Harmony Hero gameplay
Python

Python that simulates the general rules and scoring of Texas hold'em. Makes use of hierarchical class structures. Deals hands to a variable number of players, and the winner is decided based on comparative hand (card) scores. Hand scores are determined based on the actual rules of Texas Hold-em.

Chrome Extension

Productivitee is a chrome extension that allows the user to monitor exactly how long they are spending on unproductive websites. The extension can be interacted with by clicking on the icon in the top right hand corner of your chrome browser. When installed and switched on, the extension will begin tracking the amount of time the user spends on a list of "unproductive" websites.

Due to changes in JavaScript that dictate the maximum number of calls you can make to a function in a single minute of real time, the extension needs to be updated. However, here is a video of it working when first created:

03

Education

Harvard University

September 2020 - December 2024
B.A., Computer Science & Music

Concentration: Computer Science

Secondary: Music

CS124Data Structures & Algorithms CS182AI & ML CS242Advanced Machine Learning Concepts Stat110Probability CS61Systems & Machine Organization ES156Signal Processing 21M.385MIT Music Systems

Brunswick High School

Connecticut
August 2016 - May 2020
04

Skills

Programming Languages & Tools

Python Java TypeScript JavaScript C++ Node.js Flask PostgreSQL MySQL AWS Google Cloud React React Native HTML5 CSS3 Git

AI / ML

LLM agent development Guardrails & evaluation Model training RAG / tool-use integration PyTorch Generative models

Cross-Functional

Cross-team delivery Technical communication Stakeholder alignment Design review
05

Interests

Music

Apart from being incredibly passionate about music in my professional life, I also compose & produce music on Ableton Live in my free time. I have built instruments using breadboards, arduinos, and Ableton. I sang Tenor in the Harvard Callbacks a cappella group at school. I also was the Business and Tour manager of the group.

Beyond music

Apart from music, I played varsity squash, and I play squash and golf in my free time. I also enjoy Spanish and Italian, and I love indie video games like Undertale and Earthbound. Maybe one day I'll develop my own video game! (and make the music?)