Inspiration

Pockit is the expense report that pays itself. Student org members who pay out of pocket wait weeks to be repaid. Snap a receipt. AI reads it, Pockit checks your org's rules, and PayPal pays you in minutes.

Some of the most valuable learning in college happens outside the classroom: presenting at a conference, competing at a national event, running a workshop for your community. At Georgia Tech, those experiences are often powered by student organizations, and they depend on students paying up front and waiting to be paid back.

We looked around our own campus and saw that the reimbursement process is quietly keeping students from these opportunities. Corporate expense tools are built for companies with finance teams. Student orgs are run by volunteers, and the members fronting the money are students on tight budgets. Nobody was building for them.

We believe every student deserves to go to the conference, attend the competition, and take part in experiential learning, without worrying whether they'll get their money back. Pockit exists to make that happen.

What we heard: customer discovery

Before building, we interviewed five student organization leaders at Georgia Tech.

Who Role What stood out
Tristan Student org leader Reimbursements take 30–45 days. He's still waiting on one from early October.
Henry Former Secretary-General, Georgia Tech Model UN At least two former members are out thousands of dollars.
Jmar Student org leader His org stopped doing reimbursements entirely because the money comes out of members' pockets.
Kailyn Student org treasurer Tracks spending with spreadsheets, slides, and checkbook logs.
Prince RSO treasurer Had to resubmit every form because of a new signing rule.

Five patterns emerged:

  1. Students wait weeks, sometimes months. Tristan described a 30–45 day process. Henry's org used to wait months for a check in the mail.
  2. Students lose real money. Henry told us former members are out thousands of dollars. Jmar was never reimbursed for a Canva Pro subscription.
  3. Orgs avoid reimbursement altogether. Jmar's org never does reimbursements, and Tristan called them a "hail mary." When fronting money is this painful, students stop doing it, and opportunities disappear.
  4. Nobody knows the rules. Asked how he knows what's eligible, Henry said: "Frankly, I do not know." Tristan didn't know which forms he needed, and Henry's org was told flights were covered, then denied.
  5. The paperwork is brutal. Four DocuSigns per reimbursement (Jmar). Bank statements plus crumpled receipts (Prince). Resubmissions over signing rules (Prince).

What treasurers want: four of the five leaders said they'd automate the busywork but wanted the final say on larger or unexpected expenses. They worried about fraud, money going to the wrong person, and legitimate expenses being wrongly rejected. Two raised a concern we'd discover ourselves while building: not every student has a PayPal account.

Every major design decision in Pockit traces back to these conversations.

What it does

Pockit is a complete product for both sides of every reimbursement.

For treasurers

  • Add members to the org, and approve requests from students who want to join
  • Set the org's rules: spending caps by category, no alcohol, and which expenses always need review. Or upload a policy document and let AI suggest the rules, which the treasurer reviews before saving
  • Review only the exceptions: open the receipt photo when something was unclear, then approve or deny. Approving asks "Are you sure?" and lets the treasurer pay less than the full amount
  • Track every decision in a searchable, filterable audit log, and export it to CSV for student government

For members

  • See the rules up front, so nobody has to guess what's covered
  • Upload a receipt photo and confirm what the AI read before submitting
  • Get paid in minutes through PayPal for anything that follows the rules, with the exact rule that applied
  • Get a text message about every decision

How a receipt flows

  1. A member uploads a photo of a receipt.
  2. AI reads it into structured data: vendor, date, total, category, line items, and whether it includes alcohol.
  3. The member confirms the details.
  4. Pockit's rules engine checks it against the org's rules.
  5. PayPal pays the member automatically, or the receipt goes to the treasurer.
Outcome Example What happens
Follows every rule $18 lunch Paid in full, automatically
Over a category cap $60 dinner, $50 cap Pays the cap automatically and explains why
Big-ticket expense $389 hotel Sent to the treasurer for review
Clear violation Dinner with drinks Rejected instantly, with the rule shown

For capped expenses, the payout is:

$$\text{payout} = \min(\text{receipt total},\ \text{category cap})$$

Our key design decision: AI never decides to move money

The AI reads receipts. A transparent rules engine decides. Only receipts that clearly follow the rules are paid automatically, anything unusual goes to a human, and only clear violations are rejected. This directly answers what treasurers told us: they want speed and control, and they don't want an AI mistake to send money or wrongly deny it.

Try it

How we built it

We built Pockit in stages, proving the riskiest piece first.

  1. Proved money could move. Before writing any product code, we sent a test payout from a PayPal sandbox business account (the org treasury) to a sandbox personal account (a member).
  2. Split the work around a shared contract. We agreed on the exact JSON format the AI receipt reader would return, so one teammate built the AI reader while the other built the rules engine in parallel, testing it with sample receipts.
  3. Built the rules engine in plain TypeScript, so every decision is predictable and explainable.
  4. Connected the pipeline: receipt, then rules, then automatic PayPal payout, with every claim recorded in an audit log alongside its PayPal payout ID.
  5. Built treasurer review, including protection against paying the same claim twice.
  6. Designed the two experiences, one for members and one for treasurers, then added org membership, rule setup, join requests, and automatic capping.
  7. Upgraded the audit log with AG Grid and deployed to Render.

Tools we used, and why

PayPal Developer Platform (core integration)

  • Payouts API: sends every reimbursement from the org's PayPal business account to the member's PayPal account, both automatically for compliant receipts and when a treasurer approves one.
  • OAuth 2.0 (client credentials): authenticates every request server-side, so secrets never reach the browser.
  • Sandbox environment and test accounts: a business account acts as the org treasury and personal accounts act as members, so the full money flow is real and testable.
  • Payout tracking: each payout's batch ID is stored in the audit log, so treasurers can trace every dollar back to PayPal.

Why PayPal: speed is the whole point. Payouts let Pockit send money in minutes instead of 30–45 days, with no checks, no bank details to collect, and no finance office in the middle.

Anthropic Claude (receipt reading and policy understanding)

  • Claude's vision model reads receipt photos and returns structured JSON through a tool call: vendor, date, total, category, line items, and whether the receipt contains alcohol. If something is unreadable or unclear, it flags the receipt for the treasurer instead of guessing.
  • The same model reads a pasted organization policy document and suggests rules, which the treasurer reviews before anything is saved.
  • We tell it to read dates the US way (month first) and to never invent values, and the server signs what it read so the numbers can't be edited before submitting.

Why AI: receipts are messy, crumpled, and inconsistent, which is exactly the paperwork students and treasurers told us they hate. AI turns a photo into clean data in seconds. We deliberately limited it to reading, so it can't trigger payments on its own.

AG Grid (treasurer audit log)

  • Powers the audit log with sorting, filtering, quick search, and one-click CSV export.
  • Shows who decided each claim (the rules or the treasurer), which rule applied, the receipt total and the amount paid, and a link to the receipt photo. The CSV export also includes the PayPal payout ID.

Why AG Grid: our interviews showed treasurers live in spreadsheets. Kailyn tracks everything in spreadsheets and checkbook logs, and Prince's org uses itemized Excel sheets for every event. Treasurers needed a real data tool, not a static list: filters to answer "what did we spend on meals this semester?" and CSV export to hand records straight to student government. AG Grid's Community edition is also open source, matching our MIT-licensed repo.

Render (hosting)

Why Render: Pockit needs a real server, not a static site. The PayPal client secret and AI API key must stay server-side, and payouts run in server routes. Render runs our full Next.js server with environment variables kept secure, and gives judges a working live link to try Pockit without any setup.

Twilio (text messages, optional)

  • Texts the member and treasurer about each decision. Without Twilio keys, the messages appear in a "Text messages" panel in the app instead.

Also used

  • Next.js, TypeScript, and Tailwind CSS: the app, its server routes, and its design
  • GitHub: version control and our public, MIT-licensed repository
  • Claude: a coding and planning assistant during development

Challenges we ran into

  • Unclaimed payments. Our first test payout went to an email without a PayPal account, and PayPal marked it UNCLAIMED. Two interviewees raised the same concern, which shaped our roadmap.
  • Making AI trustworthy with money. We separated reading from deciding, added a confirmation step for members, and built protection so a claim can never be paid twice.
  • Rules that reflect real life. Flights and hotels are booked weeks before a trip, so date checks had to apply only to expenses like meals and rides. Over-limit expenses needed partial payment, not outright rejection.
  • Building in parallel. Splitting the AI reader and the rules engine around a shared JSON contract let us move fast, but merging our work took careful coordination.

Accomplishments that we're proud of

  • Real money movement, end to end: from a receipt photo to a PayPal payout in minutes, in the PayPal sandbox
  • A system treasurers can trust: their rules, a human in the loop for exceptions, and a full audit trail
  • Smarter than approve-or-deny: Pockit pays what the rules allow instead of rejecting an over-limit receipt outright
  • Built for an overlooked audience: student organizations, which corporate expense tools ignore
  • Grounded in real users: every major decision traces back to interviews with Georgia Tech student leaders

What we learned

Students don't need automation for its own sake; they need speed and control. Treasurers want the paperwork automated, not their judgment. Clarity matters as much as speed: much of the frustration we heard came from not knowing what was eligible in the first place. And we learned to build the riskiest part first, which made everything after it much smoother.

What's next for Pockit

  • Venmo payouts and unclaimed-payment alerts, since not every student has PayPal
  • Travel advances, paying students before a trip so nobody has to front the money at all
  • Auto-generated reimbursement forms, compiling receipts into a single submission-ready PDF, which Prince and Tristan both asked for
  • Fee handling, so orgs can cover or batch PayPal fees

Pockit was built during the PayPal AI Hackathon submission period using the free PayPal sandbox. No real money moves.

Built With

Share this project:

Updates

Submission history