Skip to content
galenry.
Work

Real constraints, measured outcomes

We are early, so this is a short list rather than a long one. Each entry states the constraint the client was under, what we shipped, and what changed, with figures we can stand behind.

AppooppanthaadiTravelBooking platformAI assistant

Self-serve booking, and an AI that isn’t allowed to guess

A women-led travel company running trips since 2016. We designed and built their platform end to end: self-serve booking for fixed-date tours, a tracked enquiry pipeline for custom packages, and Appu, an assistant that checks every figure against live data before it answers.

Appooppanthaadi homepage: a women-led group on a hillside under the headline “Your turn to go after everyone else”, with ratings and trip counts.

01Diagnose

Every trip ran through the team. An enquiry came in, someone priced it by hand, and nothing showed which enquiries had been left waiting, so the business could only grow as fast as the team could reply.

02Design

  • Self-serve booking for fixed-date tours, with seats held during checkout and released automatically if payment doesn’t complete
  • Payments and instalments through Razorpay, settled by one code path shared by checkout and every payment webhook
  • Prices and traveller details frozen at booking, so a later tour edit never changes what a customer paid
  • Cancellations with four outcomes: seat transfer, travel credit, wait for a replacement, or refund
  • Custom-package enquiries tracked against a 48-hour reply commitment, with an alert when one slips
  • 24 transactional emails, every send logged and resendable from the admin
  • A CMS the team runs themselves, covering tours, packages, offers, blog and FAQs, behind role-based access
Booking card for the Bhutan tour: ₹39,800 per adult, 13 seats available, the selected Dec 5–12 date, a guest counter, subtotal, Book Now and EMI options.
Appu chat: asked about women-only tours to Bhutan, it replies with the 8-day Thunder Dragon Kingdom tour at 39,800 per person, the same figures as the tour page.
The AI

Appu answers from the catalogue, and checks itself first

Appu is the assistant on every page. It answers questions about tours, dates and prices using only published data. Before a reply is shown, a second pass extracts every figure in it and checks each one against the data the answer was built from. If a number can’t be traced, the reply is replaced with a safe one instead of being sent.

  • Answers only from published tours, packages and policies
  • Every figure verified against source data before it is shown
  • Anything that needs a commitment, like a price or a date, goes to a person
  • Model chosen by measurement: replies in 1–4 seconds, down from 6–9
Available tour dates: a December departure card showing 13 seats, the per-person price and an EMI badge.
Departures carry live seat counts, and a held seat returns to sale on its own if payment never completes.
Request a Quote form for a Kerala package, promising a personalised quote within 48 hours.
Custom packages still go to a person, now against a 48-hour reply commitment the system tracks.

03Deliver

Timeline
Feb → Sep 2026
Payment settlement paths
1
Transactional emails
24
AI reply time
1–4 s
518

automated tests across the platform

Counted from the project’s test suite, September 2026.

DayonePrototype · pre-developmentFood productionWholesale distributionInventory

One dashboard for a bakery, its wholesale round, and its café

Dayone runs a central production kitchen, a wholesale bakery round, and a café, and was tracking orders, production and stock across Excel sheets. We prototyped the system end to end, from the owner's dashboard down to a driver confirming a delivery on their phone. Development has not started; these are the screens the build will follow.

Dayone owner dashboard: today's sales at ₹69,650, profit at ₹41,790, 711 of 720 units produced, inventory and wholesale summaries, and tomorrow's unapproved production plan.

01Diagnose

Orders, production, stock and what each wholesale customer owed were tracked across separate Excel sheets, reconciled by hand. There was no single point where the owner could see today's sales, today's production and what needed buying before it ran out.

02Design

  • An owner dashboard reading sales, profit, production and inventory off one day's data, with tomorrow's plan flagged until approved
  • Wholesale order entry pre-filled from each shop's standing order, so staff only enter the exceptions
  • Production planning that nets orders and buffer against stock on hand, then checks the result against recipes and raw-material stock before it reaches the kitchen
  • Route loading sheets generated from confirmed orders, with a running vehicle total by product
  • A delivery screen built for a phone with unreliable signal, queuing a confirmation offline and syncing when the connection returns
  • Role-based screens for kitchen, order, delivery and café staff, with costs and margins removed from the underlying data those roles receive, not just hidden on screen
Wholesale order entry for ABC Store on a phone: five products pre-filled from the standing order, quantities adjustable with plus/minus, totalling ₹905 against a ₹4,820 outstanding balance within credit limit.
Production plan for Thursday: required, buffer and final-plan quantities per product beside a material feasibility check showing maida and chocolate short against kitchen stock.
The plan is exploded against recipes and checked against stock before it reaches the kitchen, so short ingredients are flagged rather than discovered mid-batch.
Raw material stock ledger showing days of cover per item, a purchase recommendation for maida, and an open stock audit recording a −6 kg variance without overwriting system stock.
A purchase recommendation weighs stock, average use and confirmed orders. The owner still approves the number before it becomes an order.

03Deliver

8

workflows prototyped

Counted from the prototype set, September 2026. Pre-development: no build has started.

A short list, honestly presented

We would rather show work we can talk about in detail than a wall of logos we cannot. Ask us about anything here and you will get the version with the difficult parts left in.

Ask about an engagement
Start

Tell us what you’re trying to build

An idea, a stalled project, or a system that has outgrown its architecture. Send it over and we’ll tell you where we would start and whether we’re the right team for it.

Response time
Within one business day
First call
30 minutes, no deck

Prefer to write first? Send the brief and we’ll come back with questions before we come back with a proposal.