Accord

Accord is an AI group-purchasing coordinator that privately understands each member’s constraints, finds options everyone can accept, and manages consent as prices and terms change.

Comment

Inspiration

Group trips often stall for a reason people don’t want to say out loud: money. In a CIT Bank/Harris Poll survey, 82% of group travelers said they would pay more than their share to avoid money conflict and 45% had already experienced financial conflict or discomfort on a trip. The pressure doesn’t end when someone fronts the bill: Zelle research found 76% of Gen Z who covered a group expense weren’t fully repaid. And among Americans who skipped a social commitment because of cost, 56% didn’t tell loved ones that money was the reason, according to the CFP Board.

We built Accord, an AI group-purchasing coordinator that helps people find common ground without forcing anyone to disclose more than they want to.

Most AI shopping tools are built for one buyer. Accord is built for groups.

What it does

Accord helps a group plan and book a shared stay.

Each member privately describes what works for them by text or voice. Accord turns that input into structured, reviewable requirements such as:

  • maximum contribution
  • availability and trip length
  • checkout deadlines
  • refund requirements
  • accessibility needs
  • destinations to avoid
  • soft preferences like walkability or quiet neighborhoods

Members confirm those requirements before they become part of the plan.

Accord then searches real stay inventory and checks every option against the group's confirmed hard requirements. The group can see whether a stay works for everyone, but not whose private boundary caused another option to fail.

Once the group agrees on a stay, every member approves the same exact proposal and their exact contribution.

If the merchant changes the price, cancellation policy, room, availability, or another material term, Accord stops the transaction, invalidates stale approval, privately explains the issue to affected members, and replans.

A budget ceiling is not consent. Being able to afford a new price does not mean you approved it.

How we built it

Gemini

Gemini is Accord's language layer.

It converts natural conversation into structured requirements, asks targeted clarification questions, and explains options in a way that respects privacy.

For example:

"I need to leave by noon Sunday, but I don't want to explain why."

becomes a proposed hard requirement:

Checkout by Sunday at noon
Reason remains private

The member still reviews and confirms it.

Gemini helps interpret intent. It does not decide whether a stay is feasible or whether a purchase can proceed.

Deterministic constraint engine

Accord evaluates every stay using deterministic rules.

It checks:

  • personal contribution limits
  • dates and checkout times
  • cancellation terms
  • accessibility evidence
  • capacity
  • availability
  • offer expiry

Missing evidence cannot satisfy a hard requirement.

Only stays that satisfy every hard requirement are eligible for ranking.

This lets Accord use AI for human language while keeping financial and consent decisions explicit and predictable.

Backboard

Backboard gives each member persistent personal planning memory.

A member can choose to remember reusable preferences such as:

"I usually prefer walkable neighborhoods."

On a future trip, Accord can ask:

"I remember that walkability matters to you. Apply that preference here?"

The user can apply it, review it, or ignore it.

Budgets and payment permissions never become standing memories.

ElevenLabs

ElevenLabs brings voice into the same private intake flow.

Members can describe what they need naturally instead of filling out every field manually. Spoken input is transcribed, interpreted by Gemini, and shown back to the member for confirmation before it affects planning.

LiteAPI and SerpApi

LiteAPI and SerpApi connect Accord to real stay inventory.

Accord normalizes listings into one offer model containing details such as:

  • price
  • fees
  • room
  • dates
  • capacity
  • refund terms
  • accessibility evidence
  • availability
  • expiry

The deterministic engine evaluates those facts instead of trusting an AI summary of the listing.

Tiger Data

Tiger Data gives Accord a temporal view of the purchase.

Accord records public-safe events such as:

  • price observations
  • proposal creation
  • approvals
  • merchant changes
  • stale consent
  • replanning
  • booking activity

Those events power:

  • live group activity
  • price history
  • observed offer stability
  • replan latency
  • transaction timelines

This lets the group see not only what an offer looks like now, but how it has changed during the planning process.

Stability is only a soft signal. It never overrides hard feasibility.

MongoDB Atlas

MongoDB Atlas stores Accord's live operational state:

  • users
  • sessions
  • groups
  • private constraint capsules
  • offers
  • proposals
  • approvals
  • payment authorization state
  • bookings

Private data is kept behind authenticated member-specific access boundaries, and public responses are explicitly shaped so private constraints are never accidentally exposed to the group.

CyberSource

Accord binds payment authorization to the exact proposal being approved.

Before booking, Accord verifies that:

  • every active payer approved the same proposal
  • every authorization matches the correct contribution
  • the merchant quote still matches
  • the offer is still available
  • the proposal has not become stale

The backend integrates CyberSource sandbox authorization, capture, and reversal.

If the merchant changes the offer, Accord will not reuse approval for the old version.

Vultr

Vultr provides Accord’s deployment path. We packaged the API and frontend as a containerized service and configured Caddy to handle HTTPS and route traffic to the app. This gives Accord a clear path to run as one shared coordinator, so members on separate devices can collaborate in real time.

Realtime coordination

Accord uses realtime server events to keep everyone synchronized across separate browsers and devices.

The group can watch the shared transaction move through states such as:

Proposal ready
3 of 4 approved
4 of 4 approved
Merchant changed offer
Purchase paused
Replanning
New proposal ready

Private explanations are only sent to the relevant member.

Challenges we ran into

The hardest challenge was preserving privacy across the entire system.

Private information can leak through more than a database. It can leak through AI prompts, API responses, explanations, realtime updates, analytics, or even a suggestion about how the group should compromise.

We separated public and private information paths so each part of Accord receives only what it needs.

We also had to handle a harder question:

What happens when the thing everyone approved changes? Our answer was to make consent proposal-specific.

A price change, refund-policy change, room change, or other material update makes the old approval stale. Accord stops, explains the problem privately, and asks again.

Accomplishments that we're proud of

We built Accord as one coordinated system rather than a collection of AI features.

A single flow connects:

  • conversational AI
  • private personal memory
  • real stay inventory
  • deterministic constraint evaluation
  • temporal market data
  • realtime collaboration
  • proposal-specific consent
  • payment authorization

Each technology has a clear responsibility.

Gemini understands what someone means.
Backboard remembers what they want remembered.
MongoDB stores what is true now.
Tiger Data tracks what changed over time.
CyberSource handles payment authorization.

Together, they let a group move from private needs to a shared purchase without treating privacy or consent as an afterthought.

What we learned

AI is most useful when it helps people express and understand messy human requirements.

The decisions that control money and hard constraints should stay explicit and deterministic.

We also learned that memory needs boundaries. Remembering that someone prefers walkable neighborhoods is useful. Treating an old budget or payment decision as permanent permission is not.

Most importantly, we learned that consent belongs to one specific decision.

If the price, room, refund policy, or group changes, Accord should not guess.

It should ask again.

What's next for Accord

Travel is our first use case, but the coordination problem is much larger.

Accord could eventually support:

  • flights
  • event tickets
  • group gifts
  • roommate purchases
  • student organizations
  • family purchases
  • shared activities
  • team procurement

The long-term vision is to become a trusted coordination layer for AI-powered group commerce.

Each person keeps control of what they share, what they approve, and what they pay for.

Plan together. Pay together. Keep your boundaries private.

Built With

Share this project:

Updates

Submission history