Hacker
Afreen Hossain
@afreenorbit
CURIOUS
Elsewhere
Stack
- JavaScript
- Python
Hermit : A little visitor between friends
Flip the Switch HackathonNov 2026
Hermit is an animated desktop companion that visits one friend’s laptop at a time. We built it because staying in touch can be difficult even when friends live nearby. The three of us are in Bengaluru, roughly 25 km apart, but weeks can pass without a proper check-in. A group chat gives us somewhere to talk; it does not provide a natural reason to begin, and messages often arrive when someone is busy. Hermit creates a gentler opening: a small visitor appears when a friend is available and brings something from the last person it visited. - Each laptop runs its own keeper and mascot. The keeper checks local keyboard and mouse idleness and decides whether that person is available. Every mascot shows where Hermit is: visiting its current host, waiting for a friend, sleeping when nobody is around, or explaining that its connection is offline. Friends do not have to remain in a call or keep a chat window open to be part of the experience. - A visit gives people something to do together. The host can leave a small favor for whoever Hermit visits next, such as recommending a song, answering a question, or sharing a photo. The next friend receives it in the mascot’s Quest tab. Completing it adds a piece to a shared shell. Someone who is not hosting can also send a note to the current host. The shell and optional journey page turn these small exchanges into a visible history of the friendship. - Switch makes Hermit a shared experience rather than three separate desktop pets. Each keeper is registered as a custom Switch agent. The keepers use a shared ledger room to publish and receive presence updates, holder changes, favors, notes, and shell pieces. When one keeper records a change, the others update their local view and tell their mascots what to show. A favor can be left before anyone knows who will receive it; Switch carries it until the next visit. If Switch is unavailable, the laptops cannot agree on Hermit’s location or pass anything between friends, so the mascots report the connection problem instead of guessing. We host one Switch server on Afreen’s laptop. Tailscale provides a private network connection so Akshay’s and Abhishek’s laptops can reach that server. Switch carries the shared events; Tailscale connects the devices. Each mascot talks only to its own keeper through a local API, and keyboard activity is reduced to availability rather than shared as keystrokes or screen content. The keeper, animated mascot, presence views, favor relay, notes, and shared shell are implemented. Automated checks cover the workflow, and live Switch checks have shown a holder on Afreen’s desktop, an assigned favor from Akshay, and the shared shell and Friends views. Hermit’s aim is simple: make reaching out feel easier, give a brief visit a purpose, and let small acts of friendship add up over time.
3Builds@abhishiv17,@imakii
Pinlight: learn, build and battle electronics, all by voice
No Hands: The Wispr ChallengeOct 2026
When you’re new to electronics, AI tools hand you a parts list you can’t verify, and nobody’s around to check it. Pinlight is a voice-first electronics workbench for students and hobbyists: learn the part, review the idea, see it assembled, then explain it back. - Learn: 3D Raspberry Pi 5, Arduino Uno and 10 chips and sensors with 20 guided activities. Ask the tutor and the camera flies to the part. - Build Review: speak your idea, your parts and what another AI suggested. Mastra agents on Groq plan and review it, and a five-branch graph shows what you have, what’s missing, what each part adds, what happens without it, and the alternatives. - 3D build and enclosure: the reviewed build becomes a wired 3D assembly and a project-specific case, built as real solid geometry in WebAssembly (manifold-3d) and downloadable as STL or OpenSCAD. - Arena: explain concepts out loud against personality bots or a friend, graded on accuracy and clarity. Weak topics become personalised practice. Every model output passes through Zod schemas and deterministic code. The code checks wiring (missing resistors, 5 V into 3.3 V pins, missing grounds), repairs scenes (a rover always gets two drive motors on separate driver channels), packs parts without overlaps, routes wires between named pins, builds the enclosure geometry, and scores answers against fixed key points. 108 tests cover these rules. Wispr Flow is the input for everything. Dictation auto-sends when you finish, saying “that’s my answer” submits an Arena round, and a voice command bar (“open arena”, “new build”, “help”) moves you around the app. Typed by Wispr.
KindRed : When A Donor Can't Match, There's Still A Way
Graph HacksOct 2026
THE PROBLEM India meets only about 7% of its kidney transplant need and about 1.5% of its liver need. Many families have a relative who is willing to donate but can't, because the blood group doesn't match or the patient has antibodies against the donor's tissue type. A paired exchange fixes this: two or three such families swap donors so every patient gets a compatible organ. India has done only 1,839 kidney exchange transplants in 24 years and 259 liver exchanges, mostly inside a single hospital's small pool. Finding a swap is only half the problem. Every planned donation still has to pass a real test before surgery: a lab crossmatch for kidney, a CT work-up for liver. If one test in a 3-way ring fails, the whole ring collapses, and a coordinator starts again by phone, often re-planning families who were fine. WHAT KINDRED DOES KindRed is a coordination and decision-support platform for kidney and liver paired exchange across hospitals, on one FalkorDB graph. The demo pool has 198 synthetic families across 11 hospitals. 1. Match across hospitals. Each hospital keeps its private data in its own graph and shares only de-identified facts with a common exchange pool. The graph finds 2-way and 3-way swaps, and a CP-SAT optimiser picks the set with the most expected transplants (this week's plan: 32 swaps, 64.1 expected transplants). 2. Explain every match. Each possible donation stores its reasons (blood group, antibody check, tissue typing gaps, graft size for liver), so the app can say in plain words why a link works, why a family's own donor can't give, and which links are risky before any test runs. 3. Recover when a test fails. The failure is stored in the graph so that donation is never proposed again. KindRed searches only 2 hops around the break for a repair (55 ms in our run, with the other 31 swaps left untouched), keeps a 2-way fallback ready, and a human coordinator approves the fix. Nothing changes for any family until a person says yes. 4. Agents that act and remember. A coordinator agent (Groq gpt-oss-120b) runs each demo patient's case week by week. It picks one step from vetted tools over MCP (search 2-way then 3-way, form a swap, run tests, request papers, book surgery) and never writes a query. Rule-based agents (intake, matching, crossmatch, committee clerk, health watch, scheduler, patient liaison) coordinate through Task nodes in the graph instead of calling each other. Each patient's memory (findings, beliefs, playbooks) lives in its own graph. 5. Ask KindRed. Coordinators ask questions in plain language ("What breaks if this family withdraws?"). A model picks from 15 read-only graph tools, every answer shows its graph paths and tool trace, and an unknown family gets "not in the graph" instead of a guess. WHO IT'S FOR Transplant coordinators, hospital transplant teams and national or regional exchange programs, with a plain-language view for families. Walkthroughs can be read aloud in 10 Indian languages. Try it: open the live demo and press "Make my own copy" so you get a private set of graphs. Then follow Sunita (kidney) or Rajesh (liver) to transplant with Next week, or open the Live map for the failed-swap story. All data is synthetic. KindRed is a software prototype, not a clinical tool.
3 BUILDS@imakii,@abhishiv17
JalRekha: Catch lakes before they become land
Environmental HacksOct 2026
https://builder.aws.com/content/3KXo7tb9nu3AD3Hbiv0w167vPGo/lakes-keep-our-cities-cool-we-built-jalrekha-to-stop-them-becoming-land
3 BUILDS@abhishiv17,@imakii