Flight & returns
QR reading, package compatibility, booking, continuous tracking, explicit pickup confirmation, history, and simulated handoff.
CAL HACKS 13.0 · OCTOBER 23–25, 2026Four tracks,
Four tracks,
one story.
AirCart is for Berkeley residents who find getting to a return drop-off hard: people with limited mobility, older adults and car-free students. One return carries the whole demo. Claude checks it, a drone collects it, and the merchant sees what’s coming back. Everything here is labelled simulated, configured or live, and no interview, metric or sponsor eligibility is claimed.
FOUR TRACKS, ONE STORY
Live counts from /api/evidencePick your track. It’s the same return.
Each track picks up where the last one ends: the shopping problem (Visa), who it’s for and what Claude decides (Anthropic), the agent in production (Microsoft), and the drone link built to fly it (hardware). Send a judge straight to theirs with /demo#visa, #anthropic, #microsoft or #hardware.
01 · Visa · Reimagining Shopping
Shopping isn’t over until the return is.
RubricInnovation 25 · UX 25 · Technical execution 20 · Commerce impact 15 · Trust & responsibility 15 (Visa’s starter pack: UX 20 · Commerce impact 20). Return management tools are one of its example ideas.
What the demo shows
- / · customer appA return booked from home in a ride-hailing style flow: the return QR is read on the device, then a pickup spot, a drone-capacity check, a price estimate, simulated tracking (live telemetry on the drone link) and a receipt.
- Booking · Claude return checkClaude checks condition and match from a photo and suggests the merchant’s next step: restock, inspect, refurbish or return to vendor.
- /merchantThe merchant desk, a demo view of this browser’s returns, shows each incoming return with Claude’s grade and suggested disposition before the parcel arrives, not as a surprise at the counter.
- AGENTS.md · behavior contractsTrust by design: simulation is labelled, no refund, payment or retailer approval is claimed, and photos are never stored.
60-second demo path
- On /, tap Try a sample return code.
- Run the Claude return check on one item photo and read its disposition.
- Book, switch the simulator to 5× and confirm package loading.
- Open /merchant and show the return arriving with its disposition.
Live evidence
Loading live counts…
Every record in this deployment’s database counts, including team testing, smoke tests and demo-account practice. These are not customers, users or merchants. Raw counts
Honest gaps
- No payment, Visa API or refund integration. The retrieved rubric doesn’t require one; say so instead of implying it.
- Faster restock and fewer store trips are the design goal, not a measured result. No merchant has used AirCart yet.
- Hubs, prices and hand-offs are demo configurations, not participating retailers.
Keep the demo honest.
Actual build statusLive drone link
Fleet console plus a signed 4G/SMS link and a PX4/ArduPilot bridge. A full mission has run end to end with the bridge’s simulated vehicle; fly SITL before real hardware, with a pilot in command.
Open FleetAnthropic
The Claude return check, photo review and uncertainty gates are implemented. The sample output is explicitly labeled and does not establish sponsor eligibility.
Apify
Custom product-spec Actor and app integration are implemented. Missing weights or dimensions stay unknown. Deploy the Actor and show a real successful run.
Google Maps & places
Address lookup, pickup pins, UPS drop-offs and warehouse discovery are implemented. Add a restricted browser key through Maps in the tracking screen. Listings do not verify return acceptance.
Amazon returns
Profile linking needs registered Amazon app credentials. The manual Orders → QR / label → booking flow works now. Linking does not grant order history or refund access.
Continue building
The source, the Cloudflare Worker deploy guide and the drone-link protocol are in GitHub.
Browse the source How it’s deployed Drone link protocolFor the AirCart teamPitch script, rehearsal checklist, prize research and open questions. Not part of the judging.
YOUR THREE-MINUTE STORYLess running around.
Less running around.
More independence.
One return, told once, in track order: the person, Claude’s check, the flight, the merchant, and the safety close. Lead with the person; make the drone’s journey the visual payoff.
The person
Name the community: Berkeley residents with limited mobility, older adults and car-free students. Describe what a return costs them today in their words: the trip, the ride, waiting for help. Use a story the team actually heard; if nobody was interviewed, call it the problem you’re designing for.
The Claude check
In booking, describe the item and add a photo. Run the Claude return check and open its tool log: Claude calls AirCart’s drone-limit tool before it decides. Then show one case it hands to a person, and that the person can still book with their own details.
Book and fly
Book, and say whether this flight is the simulator or the drone link. Simulator: 5×, follow the drone, confirm loading, because it waits for you. Hardware: dispatch from /fleet, the operator launches each takeoff, and abort is always one tap away.
The merchant receives
Open /merchant: the return arrives with Claude’s condition and suggested disposition, so the merchant can plan to restock or inspect before the parcel lands. One sent to a person reads “Waiting for a person before it flies — inspect on arrival.” Approve it, and only then does its drone take off. Don’t quote restock-time savings you haven’t measured.
Built to be trusted
Close on architecture and safety: Worker and D1, owner-scoped records, an audit row for every Claude call, hourly caps, and signed, expiring 4G/SMS commands with the geofence and 120 m cap checked twice. End on what is live and what is simulated, using this page’s counts.
Before you pitch
Saved on this deviceEVERY TRACK, CHECKED
58 prize cards · verified Oct 23The full opportunity map.
These are opportunities, not eligibility claims. Several cards group multiple awards. Strong fit is our assessment; sponsor requirements still have to be met. Current official prize portal
| Track / prize | What it actually requires | AirCart fit |
|---|---|---|
| First OverallMacBook Air 13-inch M5 (Hive); MacBook Air or Framework per member (Devpost)ORGANIZER | Working, creative, useful project with thoughtful UX and technical depth. AirCart needs a reliable end-to-end demo and a clear target customer; prize wording differs. | Strong |
| Second OverallNintendo Switch 2 per memberORGANIZER | General organizer judging. Same AirCart submission as overall. | Strong |
| Third OverallGoogle Intelligent Eyewear per memberORGANIZER | General organizer judging. Same AirCart submission as overall. | Strong |
| Most Creative HackInstax camera plus filmORGANIZER | Original problem framing or combination of technologies. QR-to-compatible-return-route with animated drone journey is a strong visual story. | Strong |
| Most Technically Complex HackNuPhy Air60 HEORGANIZER | Demonstrate engineering depth and sound architecture. Evidence: telemetry state machine, compatibility decisions, audit events, failure recovery; animated markers alone are insufficient. | Strong |
| Best People-First HackApple Watch SE 3ORGANIZER | People-first social impact; detailed eligibility not published on Hive. Pitch a specific community with evidence, such as car-free campus residents or people with mobility constraints; show accessibility and usable fallbacks. | Strong |
| Best Planet-First HackAmerica the Beautiful Parks PassORGANIZER | Planet-first social impact; detailed rubric not published on Hive. Compare against an explicit travel baseline; label CO2 estimates and assumptions. Drone delivery is not automatically lower-emission. | Stretch |
| Best Hardware HackBambu Lab A1 MiniORGANIZER | Demonstrable hardware project; detailed criteria not published on Hive. Strong only if the drone link is shown flying a real outdoor drone (pilot in command) or a PX4/ArduPilot SITL run; map simulation alone does not demonstrate hardware. | Strong |
| Best Beginner HackAirPods 4ORGANIZER | Beginner eligibility must be confirmed with organizers. Do not infer eligibility from a first drone project; current 13.0 page does not define beginner. | Stretch |
| Best Solo HackAirPods Pro 3ORGANIZER | Solo-project eligibility. The proposal specifies four teammates. Devpost also says teams of 2–4 despite listing this award; organizer clarification needed for solo applicants. | Lower |
| Hacker's ChoiceLogitech MX Master 4ORGANIZER | Participant-voted category; voting mechanics not published on retrieved page. Polished, interactive return journey helps; do not invent voting rules. | Strong |
| Notability — Best Use of Notability1 year Pro plus merchSPONSOR | Genuine evolving build notes, clear detail, meaningful features and relevance. Keep real contemporaneous sketches, diagrams, audio and iteration notes; a retrospective token notebook is weak. | Stretch |
| Meta/Facebook — Bringing People Closer Together with AIRound 2 at Meta Menlo Park, travel/accommodation subject to rules, swagSPONSOR | Meaningful human connection, essential AI, originality and working demo. A logistics app is weak as-is; a neighborhood return-sharing/assistance network would be a larger product change. | Lower |
| Capital One — Best Use of Nessie$300 Giftogram per winning memberSPONSOR | Use Nessie; judged on completeness, creativity and complexity. Real sandbox purchase/refund audit flow fits broader plan. Current app has no Nessie integration. Clearly distinguish sandbox money from a real refund. | Stretch |
| Factory — Best Use of Droid SDK$2,000 cash plus Factory credits and swagSPONSOR | TypeScript/Python Droid SDK must do meaningful work in the running product; README explains it. Using Factory to code does not qualify. Possible future exception-resolution agent with triggers, scoped tools and visible decisions. | Lower |
| Fetch.ai — Best Use of Fetch.aiSponsor: $1,250/$750/$500 plus two $250 specials; organizer: $1,500/$1,000/$500SPONSOR | Agentverse registration, ACP, ASI:One discovery and complete primary workflow inside ASI:One; real tools/orchestration; public repo. Not satisfied by simulated drones named agents. Requires Devpost plus ASI:One submission agent, agent metadata/tags and 3–5 minute video. Payment protocol and interactive cards are bonuses. | Stretch |
| Armadin — Best Security HackApple headphones; possible hacking tools, unconfirmedSPONSOR | Criteria marked Coming soon. QR payload validation and audit logging alone do not establish security-track eligibility; wait for actual challenge. | Lower |
| Coinbase — Best Use of x402$500 USDC plus recruiter call per member per track; up to $4,000 stated for two tracksSPONSOR | Meaningful x402 use; two-track brief linked but PDF contents not yet verified. Future paid relay/service requests could fit. Do not claim a mocked payment or ordinary payment API qualifies. | Stretch |
| CREAO — Best AI-First DisruptorHive: 2,500 plus Launchpad benefits; Devpost description: $3,000 cash plus benefitsSPONSOR | Sponsor brief needed to verify exact CREAO integration requirements. AI-assisted returns and merchant operations could fit, but using Claude alone is not evidence of CREAO participation. | Stretch |
| CREAO — Best One-Person CompanyHive: 1,500 plus Launchpad benefits; Devpost description: $3,000 cash plus benefitsSPONSOR | Sponsor brief needed; title refers to business automation, solo-team rule not verified. Possible merchant operations automation; do not assume four-person team is excluded solely by category name. | Stretch |
| CREAO — Best Team EmpowermentHive: $1,000 total cash wording plus Launchpad; Devpost: $3,000 cash plus benefitsSPONSOR | Sponsor brief needed to verify exact requirement. A merchant review/exception queue could fit after real CREAO use; prize publications conflict. | Stretch |
| CRS — Financial Intelligence Challenge$1,000 Overall; $500 Financial AI Agent; interviews for Identity/Fraud and Unexpected UseSPONSOR | Meaningful live CRS sandbox API/MCP usage, central to product; provided test identities only. Return eligibility should not use customer credit. Fraud workflow would expand scope materially; avoid collecting real sensitive identities. | Lower |
| Hyphenate — Autonomous Systems for Finance$8,000/$5,000/$3,000; expedited interviews top fiveSPONSOR | Realistic messy finance problem, believable autonomous work roles, limited access, audit trail, escalation, working live demo and evidence of what is simulated. Merchant refund reconciliation plus discrepancies and approval could fit a focused finance product; simple returns booking is insufficient. Video should begin with decisive failure/recovery moment. | Stretch |
| Cartesia — Best Use of CartesiaMeta glasses, still TBDSPONSOR | Voice must be core; working demo, two-minute demo-first pitch, setup/features/limits README and source access. Hands-free accessible booking and tracking could work. Browser speech synthesis or a decorative read-aloud button is not Cartesia integration. | Stretch |
| Bitdeer AI — Best Use of Model Studio1st: $500 credits plus four gaming mice; 2nd: $200 credits plus one LEGO Technic setSPONSOR | Working Model Studio application; explain and demonstrate usage. Another model service adds scope without improving the current Claude-focused return flow. Credits are not cash. | Lower |
| Figma — Idea to DemoBrief: Design $500; Creative $500; Workflow $2,000; Stage Log $2,000; recruiter sessionSPONSOR | Public Figma files, Edu verification, <=3min demo, required Figma Slides, stage log <=200 words/stage, prototype embeds and repo. Use real iterative workflow; log ideation/prototyping/handoff/code generation, product, purpose, minutes, 1–5 rating, output. Five $1k Devpost categories conflict with four-category sponsor brief. | Strong |
| Moss — Best Use of MossAirPods Max/Pros; exact item unclearSPONSOR | Examples: live voice agent, persistent personal agent, or social-post engagement. Could add return assistant memory, but exact requirements are sparse and current app uses no Moss. | Lower |
| SpaceXAI — Make it LegendaryCustom Cursor mechanical keyboardSPONSOR | Hive: build with Cursor and use Grok Imagine or Voice API; space-data theme; Grok Bot bonus. Devpost calls it SpaceXSI and omits Grok API requirement; no current tool usage and drones are not inherently space-data projects. Separate from MLH workflow raffle. | Lower |
| Convex — Best Use of ConvexHive: $5,000 cash; Devpost: EP-133 K.O. II plus ESP32-S3 boardSPONSOR | Meaningful reactive backend usage. Excellent technical fit for shared job state and telemetry; current D1 persistence is not Convex. Choose one real primary realtime backend instead of stacking databases. | Stretch |
| Microsoft — Built for the Real WorldSurface laptop per member, FDE tech talk and mentorshipSPONSOR | Production-minded AI agent, customer understanding, measurable usefulness, production path, responsible design and end-to-end functionality. No mandatory Azure/Microsoft API stated in retrieved rubric. Show real AI review, manual escalation, audit and honest simulation boundary. | Strong |
| Microsoft — Customer Impact AwardFDE tech talk and senior-FDE mentorshipSPONSOR | Same rubric as Built for the Real World. Collect a narrow user story and measure steps/time removed; avoid invented customer metrics. | Strong |
| Apify — Best Use of Apify$2,000/$1,000/$800/$400/$400SPONSOR | Meaningful working Apify use and overall project quality. Live product URL enrichment with source URL, actual dimensions/weight, failure handling and user confirmation fits. Keyless fixture mode does not prove live integration. | Strong |
| Apify — Best Actor$400SPONSOR | Build an Actor for this event; meaningful Apify usage. Ship and demonstrate the custom product-spec Actor; an adapter alone or unrelated existing Actor does not demonstrate building an Actor. | Strong |
| Elastic — Best Use of Agent Builder on Serverless1st Meta Quest 3S; 2nd Bose QuietComfortSPONSOR | Devpost names Agent Builder on Serverless; Hive detailed criteria Coming soon. Search retailer policies/facility compatibility through real Elastic tools. Do not copy 12.0 MCP rules as confirmed 13.0 requirements. | Stretch |
| Visa — Reimagining Shopping$5,000SPONSOR | Rubric (Hive): innovation 25, UX 25, technical execution 20, commerce impact 15, trust/responsibility 15; Visa's starter pack weights UX 20 and commerce impact 20. Return management tools are a listed example idea, and Visa provides no APIs for this challenge. Returns improve post-purchase merchant/customer experience. No Visa payment integration requirement stated in the retrieved rubric. | Strong |
| AfterQueryComing soonSPONSOR | Coming soon. Track and award are listed, but criteria are not available; do not invent an integration requirement. | Lower |
| Annapurna LabsComing soonSPONSOR | Coming soon. Sponsor existence is confirmed; no specific eligible build or prize verified. | Lower |
| Anthropic$5,000 in Claude credits, 1 year of Claude Max 20x and office hours (Devpost: [Anthropic] Claude for Public Benefit)SPONSOR | Claude for Public Benefit. Named narrow community; a lived or interviewed problem; a working useful flow; meaningful Claude role with understood failures. Photo review should be advisory, flag uncertainty and require human override for uncertain package or policy details; actual Claude call needed. | Strong |
| Atlassian — Best Agent / Best Security Solution for AgentsAirPods MaxSPONSOR | Forge app for Atlassian workflows or agent governance; choose one of two tracks; <=3min demo plus repo. Real Jira/Confluence exception-workflow or agent-action audit product would qualify in principle; merely planning AirCart in Jira does not. | Lower |
| Bright Data — Best Agent/App Using Public Web Data1st $1,500 credits+$500 Amazon; 2nd $1,000+$350; 3rd $500+$250SPONSOR | Live public web data must be essential, real and robust; no mock/hardcoded demos, login/paywall scraping or private personal data. Alternative to Apify for live package specs/policies. Do not add a duplicate thin adapter solely for a second badge. | Stretch |
| Mana.spaceComing soonSPONSOR | Meaningful development use; 1–2 public Mana pages/chats, explain tool contribution and show one build moment; live app bonus. Can support a future genuine prototype/debugging stage; do not claim past work was built in Mana. | Stretch |
| SpacetimeComing soonSPONSOR | SpacetimeDB should be the core real-time backend with meaningful shared state and sync. Strong fit for synchronized 15-drone simulation or operations dashboard; alternative to Convex, not a token parallel database. | Stretch |
| Build Your Hackathon A-Team with SpaceXSICustom mechanical keyboardMLH | Cursor as core coding agent plus Grok Bot workflow automation, describe both and opt into raffle. Different from sponsor SpaceXAI API/space-data challenge; current Codex development does not establish eligibility. | Lower |
| Best Use of ElevenLabsWireless earbudsMLH | Integrate ElevenLabs audio in the working project. Accessible narrated booking/tracking could fit; browser speech does not. | Stretch |
| Best Use of Gemini APIMLH swag kitsMLH | Meaningful Gemini API application. Do not duplicate Claude image review without a user benefit; current no Gemini call. | Lower |
| Best Use of SolanaUnruggable hardware walletsMLH | Working meaningful Solana usage. Future verified relay receipts/payments possible, but unnecessary blockchain adds risk and scope to demo. | Lower |
| Best Use of Tiger DataStream Deck MiniMLH | Use Tiger Data for real data/analytics. Telemetry history and flight aggregates fit, but would add another datastore; defer unless central. | Stretch |
| Best Use of PresageFitbit Inspire plus Presage credits/discount perksMLH | Use Presage SDK human-sensing capabilities. Human physiology/emotion sensing is not necessary for returns; do not add camera health inference. | Lower |
| Best Use of Gen AIAssorted prizesMLH | Use publicly available generative-AI APIs in a novel working app. Actual Claude return-photo review naturally overlaps; demo fixtures alone are not a live API integration. | Strong |
| Best Use of BackboardTile Essentials PackMLH | Use Backboard for persistent AI context/memory and a seamless experience. Could remember return preferences across sessions; not current database-only persistence. | Stretch |
| Best Use of VultrM5Stack Tab5MLH | Build/deploy meaningfully on Vultr. Current hosting does not qualify; do not move hosting solely for a peripheral prize. | Lower |
| Commit Fellowship Prize CategoryInterview plus The Lean StartupMLH | Working team prototype, real problem, opt in and complete supplemental form. Good low-build-effort path for AirCart founders; an interview is not automatic admission. | Strong |
| Best Use of DigitalOceanRetro wireless mouseMLH | Use DigitalOcean infrastructure or Gradient AI. Current deployment is elsewhere; hosting migration not necessary for AirCart core. | Lower |
| Best Use of Snowflake APIArduino Student KitMLH | Working Snowflake API usage. Possible fleet analytics later, but redundant infrastructure now. | Lower |
| Best Use of Auth0Wireless headphonesMLH | Use Auth0 APIs for authentication/agent security. Meaningful merchant/operator identity is relevant if implemented and demonstrated; current demo lacks Auth0. | Stretch |
| Best Use of MongoDB AtlasM5Stack IoT KitMLH | Build using MongoDB Atlas. Would duplicate current backend; not automatically satisfied by generic persistence. | Lower |
| Best .Tech Domain NameDesktop microphone plus .tech domain up to 10 yearsMLH | Use an eligible .tech domain; verify exact registration rules with MLH. Branding route only if team has an eligible domain; not permission to buy one. | Stretch |
| Best Domain Name from GoDaddy RegistryDigital gift cardMLH | Register/use eligible GoDaddy Registry domain per MLH rules. Exact eligible suffix and registration process should be checked before purchasing. | Stretch |
Resolve these before submission.
Convex
Hive: $5,000 cash · Devpost: EP-133 K.O. II and ESP32-S3 touch board
Unresolved; check sponsor/organizers before promising prize.Fetch.ai
Sponsor: $1,250/$750/$500 plus two $250 special prizes · Hive and Devpost: $1,500/$1,000/$500
Unresolved.Figma
Linked sponsor brief: Four categories: $500/$500/$2,000/$2,000 · Devpost: Five $1,000 categories
Use linked brief to prepare deliverables; confirm award categories.CREAO
Hive: 2,500/1,500/1,000 descriptions plus Launchpad · Devpost: $3,000 cash description each, displayed $8,760 in cash
Do not treat displayed package values as cash; brief inaccessible.SpaceXAI/SpaceXSI
Hive: Cursor plus Grok Imagine/Voice API; Grok Bot bonus; separate MLH raffle · Devpost: Different label and incomplete requirements
Keep sponsor challenge and workflow raffle distinct.Current event
Old site: live.calhacks.io/prizes is Cal Hacks 12.0 · Current site: https://hive.hackberkeley.org/prizes
Do not apply 2025 criteria or prizes to 2026.No general sponsor-entry limit or prize-stacking rule was verified. Confirm with organizers. Additional official resources exist for Canva, Context.dev, and Greptile, but no separate prize card was found.