Work

SOLACE

Emotion-aware AI companion · AR26 HackXelerator, Mission 3 (HRI)

SOLACE detects a user's emotional state from video using in-browser computer vision, then routes that context through a four-agent backend — Emotion, Memory, Conversation, and Ethics — to generate a compassionate, tone-appropriate reply. Built for an elderly-care and mental-wellness use case: all face processing runs client-side, so no biometric data ever leaves the browser.

I was Tech Lead — built the FastAPI backend and the four-agent orchestration, integrated Groq's Llama 3.1 for the conversation agent, and deployed the API to Railway. Asna led the frontend and the animated companion face; Neha built the face-api.js emotion-detection pipeline.

~1.2s end-to-end agent latency · 0 bytes of face data ever sent to the server

• React
• FastAPI
• Groq · Llama 3.1
• face-api.js
• Vercel
• Railway

MIRAÉ LUXE

Full-stack e-commerce capstone · team of 5

A luxury beauty e-commerce platform built as a university capstone — product browsing, cart, and checkout, with an AI beauty assistant and AI-generated product imagery layered on top of a standard storefront. I coordinated a five-person team through the build.

• React
• ASP.NET Core
• PostgreSQL

BE-04 · Tasks API

FlyRank AI backend engineering internship assignment

A containerized Tasks CRUD API — FastAPI, PostgreSQL, and Redis in Docker Compose — built to demonstrate real database performance work, not just endpoint coverage. Seeded the table with 500k rows and used EXPLAIN ANALYZE to benchmark a query before and after adding an index: the plan went from a parallel sequential scan (~75ms) to an index scan (~0.1ms).

~724x query speedup, measured with EXPLAIN ANALYZE on 500k rows — not estimated

• FastAPI
• PostgreSQL
• Redis
• Docker