HotelFlow: hotels don’t need another dashboard — they need the next action

RAVI SAXENA — Sun, 20 Sep 2026 02:04:52 +0000

This is a submission for the MLH x DEV Writing Challenge

Author: Ravi Saxena (@ravi_saxena_6fd4d0ef1c9cb) — solo. No teammates.

What I Built

HotelFlow is a hotel operations engine, not a guest chatbot.

After a guest books, the hard part is everything before checkout: late arrivals, dirty rooms, a broken AC, extra towels, a VIP moved up, and two staff off the floor. That story usually lives in WhatsApp and phone calls. A manager keeps asking: what should we do right now?

HotelFlow reads one Excel workbook (rooms, staff, reservations, rules, chaos events) and builds a live queue for the next 5, 15, and 30 minutes. It scores each job (urgency + guest impact + arrival time + VIP / out-of-order), assigns people by skill and floor, and escalates if nobody is free.

The interesting piece is Event → impact → priority → assignment → resolve. Click Simulate hotel chaos and the plan rewrites itself. I did not build another PMS. I built the layer that decides what is next.

What I learned: storing “room 101 = occupied” is easy. Conflicting priorities with four people and twelve jobs is the actual product.

Demo

Live (no login): https://hotelflow-gamma.vercel.app/

Code: https://github.com/Ravi-Saxena-kronos/hotelflow

Open the site. Badge NORMAL. Demo hotel: Grand Demo Hotel, New Delhi.
Click Simulate hotel chaos.
Badge CHAOS. NOW should show VIP 303 / 105 and repairs (leak / AC). Some staff become unavailable.
Click Reset from Excel to go back to NORMAL.
Optional: type Room 302 bathroom is leaking and guest arrives at noon → Add to queue.
Mark resolved on a card to close a job.

Partner Technologies

I did not use a paid MLH perk token in this build.

Stack: Next.js 14, TypeScript, Tailwind, Vercel, and xlsx to import the hotel workbook. The decision engine is in src/lib/engine.ts — not an LLM chatbot.

Vercel is serverless, so chaos / resolve / typed incidents send the current hotel state with the request. The default hotel is the Excel file in the repo (data/HotelFlow-input-template.xlsx).

Hackathon Experience

Solo build. I am the only author.

I built HotelFlow as an operations control room for a problem I see in hotels: the desk is not short on screens; it is short on one next action when everything breaks at once.

The part I will remember is the chaos button. A calm map is not a demo. Injecting the Excel chaos playbook (checkouts, AC, leak, VIP moved to 11:00, two staff unavailable) and watching NOW / 15 / 30 reorder is the story.

TrueStay: an agent that adds the resort fee search results hide

RAVI SAXENA — Sat, 19 Sep 2026 16:01:56 +0000

This is a submission for the Sanity Challenge, Path One: Ship an Agent That Queries Real Content

Author: Ravi Saxena (@ravi_saxena_6fd4d0ef1c9cb) — solo submission. No teammates.

What I Built

TrueStay is a resort-fee desk, not a hotel blog.

A guest types one question. The desk answers what is mandatory on top of the room rate, tonight, for this guest. It does that by loading dated feeClaim documents from Sanity and applying waiver rules. It does not search the web.

Demo hotel: Bellagio Las Vegas.

Search fails here because three official pages stay live at the same time:

The booking calendar says the shown rate excludes the daily resort fee, next to “no hidden fees.”
The property FAQ treats a daily resort fee as a real guest cost.
MGM Rewards says the fee is waived for Gold+ only when the stay is booked direct.

Google that mix and you get $55, $0, and “no hidden fees” in one blob. TrueStay stores each sentence as its own claim (amount, mandatory, effectiveFrom / effectiveUntil, sourceUrl, exact quote). GROQ loads claims that are still in force. The desk then:

adds the resort fee unless Gold+ and booked direct
adds self-parking only if the guest said they will park
never invents a room rate
shows both source quotes when two current official claims disagree

If you delete amount, mandatory, or effectiveFrom from the documents, the desk cannot answer. That is the Path one test.

Demo

Live desk: https://truestay-five.vercel.app

Studio (same project): https://truestay-five.vercel.app/studio

Judge question (leave this as-is, then click Ask with GROQ):

I'm not MGM Rewards Gold. One night at Bellagio Las Vegas, I will self-park. What is mandatory on top of the room rate?

Expected:

Extra	Amount	Why
Daily resort fee	$55 / night	Guest is not Gold+. Waiver does not apply.
Self-parking	$20 / 24h	Guest said they will self-park. Not inside the resort fee.
Mandatory extras total	$75	Room rate is still whatever the calendar shows.

Then click Gold+ guest and Ask again:

I am MGM Rewards Gold. One night at Bellagio Las Vegas, booked direct, I will self-park. What is mandatory on top of the room rate?

Expected: resort fee $0 (waiver + direct). Parking stays $20.

Conflict cards still show the calendar quote vs the FAQ vs the FTC total-price example vs the Gold waiver. Both sides stay visible. The desk does not pick a winner by keyword.

No login is required on the public desk.

Code

GitHub: https://github.com/Ravi-Saxena-kronos/truestay

Stack: Next.js 14 App Router, next-sanity, embedded Studio at /studio, one POST route at /api/ask.

The important files:

src/sanity/schemaTypes/hotel.ts and feeClaim.ts — the model
src/sanity/lib/queries.ts — current-claims GROQ
src/app/api/ask/route.ts — parse the guest, fetch claims, refuse if the dataset is empty
src/lib/solveStay.ts — Gold+ / direct / parking rules
scripts/seed.mjs — Bellagio + 5 official claims
docs/sanity-context-mcp.md — how to point a hosted agent at the same content through Sanity Context

.env.local is gitignored. Write tokens are not in this repo and not in this post.

How I Used Sanity

I modeled the problem as atomic claims, not pages.

hotel is a name, brand, city, official URL.

feeClaim is one dated fact about that hotel:

feeType — resort, self-park, waiver, advertised-rate-excludes-fees, or law-total-price
amount, currency, per
mandatory, includedInAdvertisedRate
bookingChannel (direct / ota / any)
minLoyaltyTier
effectiveFrom / effectiveUntil
sourceName, sourceUrl, quote, observedOn

GROQ used by the desk:

*[_type == "feeClaim"
  && (
    hotel._ref == $hotelId
    || hotel->name == $hotel
    || hotel->name match $hotelPattern
  )
  && (!defined(effectiveUntil) || effectiveUntil >= $stayDate)
] | order(feeType asc) {
  _id, feeType, amount, currency, per, mandatory,
  includedInAdvertisedRate, bookingChannel, minLoyaltyTier,
  condition, effectiveFrom, effectiveUntil,
  sourceName, sourceUrl, quote,
  "hotelName": hotel->name
}

That query is why this is not search. Search can find the words “Gold” and “$55” on the same page. It cannot filter “still in force on the stay date,” follow hotel._ref, or apply “waiver only if Gold+ and booked direct.”

Seeded Knowledge Base sources (re-checked against the live URL on ingest day 19 Sep 2026):

The same URLs are the ingest list for a Sanity Knowledge Base / Context MCP endpoint (initial_context, then groq_query with the filter above). The public judge demo is the GROQ desk so anyone can click Ask without installing an MCP client.

Honest limit: fee amounts can change. The desk answers from the dated quote we stored, not from a live scrape. If MGM updates the FAQ, edit the claim in Studio. That is the point of structured content.

Sanity Project Details

Project ID: 7i0vhp5u
Dataset: production (public so judges can inspect documents)
Types: hotel, feeClaim
Studio: https://truestay-five.vercel.app/studio
Live desk: https://truestay-five.vercel.app

One hotel document (hotel.bellagio) and five feeClaim documents:

`_id`	Type	Amount
`claim.bellagio.resort`	resort	55 / night
`claim.bellagio.waiver`	waiver	0 (Gold+, direct, max 2 rooms)
`claim.bellagio.parking`	parking_self	20 / 24h
`claim.bellagio.calendar`	advertised_rate_excludes_fees	calendar is room-only
`claim.ftc.total`	law_total_price	advertised total must include mandatory fees

Agent Session

Optional native upload is not attached yet. Design thread (no tokens):

User: Path one hotel topic that wins — structured claims, not search.

Agent: Official sources disagree on resort fees. Calendar excludes the fee, FAQ treats it as a cost, Gold+ waiver is only for direct booking. Demo: Bellagio. Schema: hotel + feeClaim with amount, dates, sourceUrl, quote.

User: Scaffold it.

Agent: Seeded resort $55, waiver $0 Gold+ direct, self-park $20. Non-Gold + park → $55+$20. Gold+ → resort $0, parking $20. Delete amount / mandatory / effectiveFrom and the desk cannot answer.

DEV Community: RAVI SAXENA

HotelFlow: hotels don’t need another dashboard — they need the next action

What I Built

Demo

Partner Technologies

Hackathon Experience

TrueStay: an agent that adds the resort fee search results hide

What I Built

Demo

Code

How I Used Sanity

Sanity Project Details

Agent Session