How Provis built Gyanee.ai, an AI shopping assistant that finds the best deal, coupon, card offer and cashback in one chat.
Gyanee.ai is a free assistant for Indian online shoppers. They ask by text or voice and get the best price, coupons, card offers and cashback in one place. Provis built the Flutter app, the AI backend, the website and the admin tools.
- Flutter app
- AI backend
- Website
- Admin panel
Ask one question, by text or voice, and get the best deal.
From the Gyanee.ai app, its app store pages and its offer data.
- 9
- languages
- Including English, Hindi and Hinglish, so shoppers can ask the way they talk
- 119/119
- SBI card offers now work
- Before the fix, none of them could be shown to shoppers. Now all of them can.
- 3,056
- card offers
- From HDFC, SBI, Axis and Kotak, as of October 2026
At a glance
- Problem
Finding every coupon, card offer and cashback for one purchase took too many sites, so shoppers gave up and paid full price.
- Solution
A chat assistant, by text or voice, that finds deals, coupons and card offers from one question and knows which of the shopper's own cards to use.
- Result
An app on Google Play and the App Store, with a website, web chat, an admin panel and a sales-agent panel, working in 9 languages.
- Product
- Gyanee.ai, an AI shopping assistant
- Industry
- Consumer apps (shopping): the best price, coupons, card offers and cashback
- Market
- India
- Platform
- Android and iPhone app, website with web chat, admin panel and a panel for sales agents
- Services
- App development, AI development, website, admin tools, deal and offer data
- Languages
- English, Hindi, Hinglish, Bengali, Gujarati, Kannada, Marathi, Tamil, Telugu
- Built with
- Flutter; FastAPI on Python 3.12, PostgreSQL with pgvector, Redis; LangChain and LangGraph calling AI models through OpenRouter; Sarvam AI for voice; Next.js for the website and panels; headless WordPress for the blog; Docker Compose behind nginx
- Connected tools
- Cuelinks (deals, coupons, cashback), GyFTR vouchers, MSG91 (OTP over WhatsApp), Sign in with Apple, PostHog
Saving money on one purchase meant five separate jobs.
Gyanee.ai is a free assistant for Indian online shoppers. It earns affiliate commission when a user buys through a tracked link.
This is how a shopper looked for savings before, from the product's own overview. It says most people give up halfway and pay more.
How a shopper looked for savings
- Compare priceson Amazon, Flipkart, Myntra and more
- Search for coupon codesmany of them expired
- Remember the cardwhich credit card has a discount this week
- Check a gift voucherto see if it saves more
- Log into a cashback siteseparately
Gyanee.ai does all five from one question.
Finding deals was easy. Showing only deals that are true was the hard part.
Four problems Provis found while real people used the app, and how each one was fixed.
-
Answers took 15 to 25 seconds to appear
The app waited for the full answer before showing anything, so people looked at a blank screen for up to 25 seconds.
What Provis builtAnswers now appear as they are writtenThe reply shows word by word, then the deal cards. While the app searches, a short message ("Searching for deals…") is shown. If the connection drops, the app reloads the conversation, and older versions of the app keep working.
-
SBI card offers were saved but never shown
Provis had saved 119 SBI card offers, but every one was missing its dates, so the app hid all of them.
What Provis builtOffers read from the right page, with their datesThe offers are now read from SBI's main offers page, which lists the dates, eligible cards and category of each offer. All 119 offers now show with the right dates, and loading them takes one page instead of about 120.
-
The app gave the wrong card advice for a shop
A tester with an Axis Bank Flipkart card asked which card is best on Flipkart. The app said 1% cashback, but that card gives 5% on Flipkart.
What Provis builtCard advice that knows each shop's offerThe app now checks what each card gives at the shop the user names, and lists the user's own cards first. A cheaper AI model was also tried and gave less honest answers, so the stronger model was kept, with a check that stops it being swapped by mistake.
-
Deals from Croma could not be collected
Croma had more deals than any other shop (36), but its website blocked the tool that collects them.
What Provis builtCroma deals are collected againProvis tested different ways of reading the site and moved to the one that worked. A new check also makes sure every shop that is switched on has search words, so no shop is missed again. Croma card offers now show in the app.
Built in four phases, from login and chat to memory and card offers.
- Phase 1, April to May 2026OTP login, semantic search, chat, deal cards, admin panel, website and the sales-agent panel; the app started on 20 May
- Phase 2, June to July 2026Streaming answers, an AI post generator, OpenRouter, a LangChain agent, Cuelinks and GyFTR, credit cards in chat, voice calls, web search and location
- Phase 3, August to September 2026AI cost dashboards, Sign in with Apple, campaigns with points, the memory engine, iOS live on 16 September, a blog, Add Your Cards, statement upload, Croma, SBI and HDFC offers
- Phase 4, October 2026A credit card banner on the app home, App Store version 2.0.3 on 8 October, local and radius search
-
What was built: a Flutter app for Android and iPhone, an AI backend on PostgreSQL, a website with web chat, an admin panel and a sales-agent panel.
-
Deals and cards: deals, coupons and cashback from Cuelinks, GyFTR vouchers, and a card catalogue merged from five sources.
-
Hardest part: savings claims that are never wrong. When unsure, the team dropped a claim rather than widen it.
-
Changed on the way: from Azure OpenAI to OpenRouter, and from LiteLLM to LangChain.
AI in the product:
An AI agent with 18 tools finds the deal.
It answers by text or voice, and live voice calls work in Indian languages. Evaluation sets score its answers before a change ships.
Exact card-name matches, expiry checks on every read, and web results marked unverified.
A one-time consent screen names the AI services the app uses.
A live voice call pauses when the app leaves the screen.
How Provis builds AI into a productWhat a user sees in Gyanee.ai.
Screens from the Android app, version 2.0.3. Select one to enlarge it.
-

Ask onceBy text or voice, with your cards at hand
-

Start with a promptDeals, phones, food, cards and fashion
-

It asks firstShort questions to narrow the answer
-

Deals with datesOffers with expiry dates and a link to buy
-

Your cardsSave cards and check a statement
On Google Play and the App Store, and still being built.
Gyanee.ai is live on the web, Google Play and the App Store, and Provis keeps adding to it. The figures below are from its own offer data, as of October 2026.
- 3,056
- card offers
- 642
- cards covered
- 1,100+
- merchants
- 4,600+
- coupons
- 4
- banks' offers loaded
- 11
- voice languages
After launch
Provis is still working on it.
The latest work at the time of writing is from 14 October 2026, on local search.
Added since launch: memory, Add Your Cards, a statement check, card offers, a blog and local search.
Provis has delivered 200+ projects since 2012, with a team of 25+ engineers, designers, QA and project leads.
Every Provis build comes with 24/7 monitoring and support after launch.
How every Provis project runs
- One project leadyour single point of contact
- A written weekly update and a demoof what was built that week
- A staging link and your repoyou see the work as it is built; the code goes straight into your repo
- QA before launchour QA team tests every feature
- Yours from day onecode repo, cloud account and domain in your company's name; our team joins as collaborators, and you can remove access at any time
- Our team works in India, on ISTUTC+5:30: 1.5 hours ahead of Dubai, 4.5 ahead of London (5.5 in winter), 9.5 ahead of New York (10.5 in winter)
Want to build an AI product of your own?
Tell us what you want it to make or do, and who for. We will come back within two working days with an approach, a rough timeline and a rough cost range. No charge, no commitment.
Send us a message
indicates a required field