Ananya Jajoo
I’m a Computer Science student at Purdue University, interested in AI systems, agents, cloud infrastructure, simulations, and developer tools.
Experience
Built MCP and agent tooling that parses internal wikis and live tickets, improving domain-specific query correctness from 13% to 85%, and developed a root-cause analysis agent that shortened Sev-2 mitigation from hours to minutes.
Python · TypeScript · MCP · Claude Code · Amazon Kiro
Built a distributed RAG pipeline on AWS Bedrock and Claude that reduced internal document search time by 97%, and trained N-BEATS models that beat XGBoost and linear regression baselines for ceramic yield forecasting.
AWS Bedrock · Claude · N-BEATS · SageMaker · Lambda · Redshift · GitHub Actions
TODO: Add a 1–2 sentence description of this internship from your resume or notes.
Built 30+ reusable React components and a Tools App that helped 50+ engineers adopt a new component library and design standards, informed by UX tests and VOC interviews.
React · UX research
Integrated 35+ C and Java features across enterprise applications, querying complex financial data in PostgreSQL and shipping production PRs with unit tests in Agile sprints.
C · Java · PostgreSQL
Built jsPsych web experiments for psychology data collection and an OpenAI-powered autograder that sped up feedback for 80+ participants.
jsPsych · OpenAI
Projects
An agentic simulation of human behavior: “Simulate Anything.”
There is no easy way to model a realistic human social network and study information diffusion without space, time, and ethical constraints. Polling is slow and surface-level, and focus groups cannot capture emergent dynamics across thousands of interconnected people.
Engineered an agentic simulation with 50+ DeepSeek-R1 (32B) agents on Modal, with persistent belief-shift memory and 26 KB isolated BFS context routing. A planner model grounds digital-twin agents in census data, then runs short- and long-term information diffusion simulations that researchers can interview directly. Stack includes FastAPI, React, and Three.js.
Binge markets — a Netflix-style interface for prediction markets.
Prediction-market platforms carry high cognitive friction for new traders, which keeps casual participants out of a market where recreational traders already drive a quarter of activity.
Built a React frontend over a Node/Express API serving live Polymarket Gamma and CLOB data, plus a Python/FastAPI recommendation service. The recommender combines NLP bigram analysis for topic clusters, multi-factor scoring on volume and novelty, temporal awareness for markets nearing their end date, and negative preference filtering. Gemini and the YouTube Embed API personalize each market view.
Redefining the human connection landscape.
Omegle made random connection with strangers compelling but shut down because an anonymous, unmoderated platform became unsafe and legally unsustainable.
Built a voice-first speed-dating app where an AI mediator joins every conversation to guide it through curated questions and keep it accountable. Three.js flame avatars grow and change color as the conversation deepens, and a face reveal unlocks only once both people reach the top level. React and Tailwind on the frontend, with Janitor AI, Claude, and Fish Audio powering the mediator.
AI security reimagined.
Airports, stadiums, and transit hubs keep growing, but surveillance still depends on human attention across hours of footage and dozens of screens, so critical moments get missed.
Built a computer-vision platform that uses vision-language models to detect people collapsing, panic-induced running, unattended objects, and dense crowd formations as they happen. Operators can query past footage in plain language (“person in red jacket running near Gate 3”) and receive email alerts. Runs on AWS with Groq, OpenAI, Roboflow, and Whisper.
Reimagining ads with interactive AR avatars so people can meet and talk with celebrity personas.
Ads are static and impersonal interruptions; slynk explores a more immersive, conversational discovery experience.
Built an AR app where scanning an ad brings its spokesperson to life in real time to walk through the product, answer questions, and show how it would look in your own space.
Winner — Sync: Best Lipsync Hack
Listed on Devpost as Anim-Education: convert live lectures into generated animations.
Traditional lectures are hard to follow for neurodivergent students, and static slides do little to make abstract concepts visual in real time.
Built a real-time pipeline converting live audio into Manim animations with Groq/NLP, decoupling generation from rendering via WebSockets, async queues, and multithreading for concurrent VR clients. Caching and buffer management keep transitions near-instantaneous.
More projects
A CPR companion on your wrist.
People freeze during sudden cardiac arrest and rarely know the correct compression rate or duration for the person in front of them.
Built a watchOS app in Swift that adapts to adult, child, or infant CPR, then drives a ring timer and per-compression haptic vibration to hold the right rate. A companion Next.js site with Firebase and Chart.js lets medical professionals review and export session data as CSV.
Winner
Reducing food waste one store at a time.
Supermarkets waste 43 billion pounds of food a year, and half of what they throw out is still edible.
Built a two-sided app: grocery stores list items nearing expiry with quantity and shelf life, and food pantries or individuals reserve them for a scheduled free pickup. Requested quantities are removed from the shared database automatically so both sides see live availability.
Winner
AI and game-like challenges as a nest for mental health.
Working and studying online removes the daily contact with friends and coworkers that keeps people feeling connected.
Built a Swift app that serves one small daily challenge at a time. The user completes it, photographs the result, and a machine learning model verifies the photo before awarding points. Firebase handles authentication and per-user progress.
Winner — Best App (StarHacks); Winner — Best High School Hack, BU Spark! (TechTogether Seattle)
Nutrify wasted food and help protect the environment.
The US produces roughly 80 billion pounds of food waste a year, much of it from households that simply do not know what to cook with what they have.
Built a mobile app whose ML model scans a food item, then surfaces its ingredients, servings, and step-by-step recipes alongside an AR scene and a page showing the environmental impact of what would otherwise be thrown away.
A resource site to expose more girls to STEM and encourage entering the field.
The myth that boys are innately better at math produces a confidence gap, even though research shows no difference in mathematical skill between genders.
Built a website compiling STEM resources into browsable sections for events, camps, mentorships, hackathons, and internships, so a student can find a next step in one place.
An integrated approach to help passengers improve their flight experience.
Passengers have no visibility into where their suitcase is until it appears on the belt, and no information about an aircraft's actual state while a flight is delayed.
Built an iOS app pairing an AR flight path and live aircraft speed, altitude, and heading with joint luggage tracking over GPS, Bluetooth, and camera.
Factual offline survival app for medical, shelter, food, and navigation information.
In a medical or survival emergency, searching the web returns hundreds of conflicting results at exactly the moment panic makes them hardest to judge — and often without signal.
Built a Flutter and Dart app with step-by-step guides for shelter, navigation, and medical scenarios, plus a machine learning model that identifies berries through the camera and reports whether they are edible. Every feature works offline.
Transport and access medical records with ease.
Patients must carry large volumes of printed exam results to every appointment, and that paperwork is easily lost to storms, floods, and other calamities.
Built an Angular web app on Firebase storing patient records in the cloud, accessible only to the patient, shareable with a hospital by consent, and reachable by a hospital directly in an emergency.
Winner — Top 20 teams
Determines who may enter an online class call using face recognition.
Zoom bombing disrupted online classes worldwide because there was no safe way to hand out a meeting link that only enrolled students could use.
Built a site using face-api.js that opens the student's webcam, detects a face, and matches it against a database of registered students, redirecting to the meeting link only on a match and labelling anyone else as unknown.
Winner
A website for a moment of peace: music, aesthetic backgrounds, and a chat bot.
Everyday life is relentlessly stressful, and there are few places online built simply to let someone step away from it for a few minutes.
Built a site in JavaScript, HTML, and CSS combining a chatbot to talk through how you are feeling, an embedded Spotify player of relaxing nature sounds, and a parallax home page of calming landscapes with overlaid quotes.
Research / Contract Projects
- Duality LabResearch AssistantAug 2025 – Dec 2025
Worked on pre-trained models for more efficient energy consumption, including the SysLLMatic project.
- AutoIC LabResearch AssistantAug 2025 – Dec 2025
Created a LiDAR 3D virtual simulation of a crane environment.
- Barrios TechnologyUnreal Engine/C++ Contract ProjectJan 2025 – May 2025
Engineered a C++/Unreal Engine VR visualization of NASA Artemis Gateway hardware, simulating modular components, docking configurations, and dynamic solar array rotation.
Activities
Hackathons
- HackIllinois2026
Matrix
- InnovateHer2026
Winner — [MLH] Best Use of Gemini API · Illume
- NexHacks2026
Polyflix
- Cal Hacks 12.02025
Flame
- Catapult Hacks2025
Spectra
- BoilerMake XII2025
Winner — Sync: Best Lipsync Hack · slynk
- TreeHacks2025
Immersive Ed
Events & competitions
- Y Combinator Startup School2026
- Crowe National Case Competition2024
4th place
- Bloomberg Code to CareerSummer 2026