Prize Winners 2026-2027

Fair Diffusion studies how retrieval-augmented generation can trace the images a model draws on and compensate the artists behind them. The project builds a reference that links a generated picture to source work, so image models can credit and pay creators instead of using their art without acknowledgment.

This project designs AI tools around the people they are meant to serve, with economic justice as the goal. It looks at how interfaces, data, and automated decisions can widen or close gaps in work, credit, and public services, and prototypes systems that keep human judgment in the loop.

Adaptive learning through computer vision uses a camera to follow how a student works and adjust instruction in real time. The system reads visual cues of progress, confusion, or engagement and changes the next exercise, so practice stays matched to the learner.

Mensajeros del Mundo is an international pen-pal platform that pairs students across countries for ongoing written exchange. The project studies how a simple online correspondence can build language practice, cultural understanding, and friendships that a classroom alone does not reach.

This project grounds large language model answers in a formal ontology, so reasoning follows defined concepts and relations instead of free-form guesswork. The aim is a model that can show which facts it used and stay consistent when a question depends on structured knowledge.

Carrier-grade NAT lets many households share one public address, and it can block or slow connections that start from outside. This project measures that inbound penalty on current protocols, documenting where CGN breaks peer-to-peer apps, games, and self-hosted services.

Finding and fixing software vulnerabilities is slow and expensive. This project works on AI agents that can locate flaws and propose repairs at a cost small teams can afford, testing whether automated remediation can stay accurate without a large security staff.

Wildfire smoke now reaches people far from the burn. This project maps who inhales that smoke, combining climate, fire, and population data to show exposure over time and where the health burden falls hardest as fire seasons grow.

Large models are costly to store and run. Smarter Compression looks for ways to shrink them while keeping the judgments that matter, so capable models can run on smaller machines and use less energy.

AgentWatch is a monitored browser sandbox for AI agents. It lets an agent use the web inside a contained session, records what it opens and clicks, and gives a person a way to review or stop an action before it leaves the sandbox.