Building reliable AI for the real world▮
I'm Summer Pandey, a Computer Science and Data Science student at Augustana College. I build human-centered AI products across edge computer vision, voice automation, and personal health. I care about privacy, reliability, and turning complex technology into tools people can actually use.
Welcome to my
room of ideas
My Work.
A collection of things I've built — products and experiments across edge AI, health technology, automation, and applied machine learning. Filter by what you're curious about; each card links out to its code, write-up, or résumé entry.
Selected work
Privacy-first edge AI for monitoring operating-room safety without sending sensitive video to the cloud.
Operating-room safety procedures must be monitored consistently, but uploading medical footage to external servers introduces serious privacy concerns.
I prototyped a computer-vision system that processes video locally on an NVIDIA Jetson. It monitors four safety workflows: hand hygiene, instrument counts, zone tracking, and sterile-field alerts.
I combined AI detection with deterministic safety rules and a human-in-the-loop review step so uncertain events are reviewed before being reported.
The prototype demonstrates how hospitals could automate safety monitoring while keeping sensitive footage on the local device.
Python · NVIDIA Jetson · Computer Vision · Gemini AI
An AI-assisted health and workout tracker that turns photo, voice, and text logs into structured daily records.
Health tracking becomes difficult to maintain when users must manually format every workout, meal, or wellness entry.
I built a progressive web app that accepts photo, voice, and text input. AI extraction and TypeScript validation transform all three input formats into consistent health records.
I created a React dashboard that combines health and workout metrics into one daily view. I also designed a PostgreSQL schema supporting five record types and used Supabase row-level security to isolate each user's data.
VentureGain supported more than 20 active users while making daily logging faster and more flexible.
TypeScript · React · Supabase · PostgreSQL · Vercel
A Gemini-powered preventive-health app that creates personalized health checklists from demographic and regional risk factors.
Generic health recommendations often fail to account for a person's demographic background, location, and individual risk factors.
I developed a Flutter application that uses Gemini AI to generate personalized preventive-health checklists based on demographic and regional information.
I added Firebase-powered progress tracking, gamification, and leaderboards to encourage users to complete preventive-health actions consistently.
Preventia won Best Use of Gemini AI at HackAugie.
Flutter · Firebase · Gemini AI
A machine-learning pipeline that scores sentiment in cryptocurrency news and social-media discussions.
Cryptocurrency discussions move quickly across news and social platforms, making overall market sentiment difficult to evaluate manually.
I developed a Python pipeline that processes cryptocurrency text and combines RoBERTa-based language understanding with VADER sentiment scoring.
The model achieved 72% accuracy when evaluated against labeled sentiment data.
The project converts large amounts of unstructured crypto discussion into sentiment signals that can be analyzed more efficiently.
Python · RoBERTa · VADER · Natural Language Processing
A Raspberry Pi robotics car that uses real-time lane detection for autonomous navigation.
An autonomous vehicle must interpret visual road information and make navigation decisions with limited on-device computing power.
I created a Raspberry Pi-powered robotic car that analyzes camera input and performs real-time lane detection for autonomous navigation.
The project focused on connecting computer-vision output to physical steering behavior while operating within the Raspberry Pi's hardware constraints.
The completed prototype demonstrated on-device visual perception and autonomous lane-following behavior.
Python · OpenCV · Raspberry Pi · Computer Vision
A Chrome extension that logs workouts from plain language, cutting manual entry time.
A Google Chrome extension to log workouts from plain language — cut manual entry 40%, 200+ entries stored.
Chrome Extension · OpenAI · Node.js
This site — a moody, film-grain, scroll-to-grow portfolio built with Next.js & React.
Twitter sentiment analysis — cleaning, modeling, and visualizing public sentiment from tweet data.
Twitter sentiment analysis — cleaning, modeling and visualizing public sentiment from tweet data.
Python · NLP · Data Science
A streaming app with autoplay channels and personalized recommendations, built for the CS SI course.
AuraTV — a streaming app with autoplay channels and personalized recommendations, built for the CS SI course.
Flutter · Firebase · YouTube API
A volleyball tournament organizer — building brackets, scheduling matches, and tracking results.
A volleyball tournament organizer — building brackets, scheduling matches and tracking results.
App · Scheduling
Multi-marketing modeling — quantifying how marketing channels drive outcomes with SQL and Python.
Multi-marketing modeling — quantifying how marketing channels drive outcomes, with SQL & Python and channel-ROI reporting.
Python · SQL · Marketing Analytics