All Projects
AI Platform / Event Tech 10 Days · Feb 2026 ● Live

Every Exhibition Booth Has the Same Problem. I Fixed It.

How I walked the floor at India AI Impact Summit 2026 — 840 exhibitors, 300,000 attendees, PM Modi inaugurating — and saw every single booth doing lead capture with zip-lock bags. Built the fix in 10 days.

<30s Card to CRM
95%+ OCR Accuracy
10 days Idea to Production
3 in 1 Bot + API + Dashboard
☕ The Order

The ₹8 Lakh Black Hole

February 2026. India AI Impact Summit at Bharat Mandapam, New Delhi — the biggest AI event India has ever hosted. 300,000 registered attendees. 840+ exhibitors across 70,000 square metres. Inaugurated by PM Modi. 118 countries represented.

I'm walking the expo floor. Cutting-edge AI everywhere — autonomous systems, LLMs, computer vision, robotics. And at every single booth, the same scene: a sales rep, a bowl of visiting cards, and a zip-lock bag getting fatter by the hour.

The irony hit hard. We're at a summit about the future of AI, and lead capture hasn't evolved since 1995.

This wasn't new to me. At Fourzip, we did 20+ trade shows selling fleet tracking hardware. At GoMechanic, at ZYOD manufacturing expos. Every single time — the same chaos. But walking that floor at Bharat Mandapam made the math impossible to ignore:

840 exhibitors × 100-500 cards each × 5 days = potentially 400,000+ business cards at a single event. And I knew what would happen to 90% of them.

The Death of a Lead

📇 Card collected Day 0
👖 Pocket / bowl Day 0
🛍️ Zip-lock bag Day 1-3
🗄️ Desk drawer Week 1
📊 Excel typing Week 2
📧 Generic email Week 3
💀 Lead is dead Week 4

Four People, Same Pain

🧑‍💼 The Booth Rep

80 cards in his pocket, can't remember who said what. By Day 2, context from Day 1 conversations is gone. The ₹50L buyer's card is somewhere in a zip-lock bag.

Deals lost to forgotten context
👩‍💼 The Sales Manager

Blind during the event. Gets a messy Excel 2 weeks later. Can't tell if team worked hard or scrolled Instagram at the booth.

No ROI data, no accountability
🏭 The Business Owner

Met the FirstCry procurement head personally. Scribbled notes on the back of the card. Assistant can't read his handwriting.

High-value relationships lost
🎪 The Event Organizer

Exhibitors keep asking 'how many leads will I get?' She has no answer. Competitor events offer lead retrieval as a perk.

Can't justify stall pricing

Companies spend ₹2-10 lakhs per exhibition stall. Travel, booth rental, salaries, collateral. Then they capture leads in a bowl and follow up 3 weeks later with a generic "nice meeting you" email.

The problem isn't laziness. It's that the process is so painful, people avoid it. Manual data entry of 200 business cards is punishment duty. And by the time it's done, every lead is cold.

I walked out of Bharat Mandapam and started building LeadSnap that night.

🔥 The Brew

Not a Card Scanner. A Lead Intelligence Platform.

The obvious solution is "an app that scans cards." Everyone builds that — CamCard, ScanBiz, HubSpot Scanner. They all miss the point. The problem isn't scanning. It's everything that happens after the scan — or doesn't.

The Pipeline

📸
Snap Photo in Telegram. That's the entire UI. No app download, no login, no onboarding.
🤖
Extract GPT-4o Vision reads the card — 14 fields including name, company, designation, phone, email, website, services, industry. Handles Hindi, Gujarati, mixed-language. Works on badges, slides, banners — not just cards.
🔍
Deduplicate Fuzzy matching on phone, email, and name+company. Catches duplicates across team members. Merges front + back sides automatically.
🏷️
Tag One-tap: 🔥 Hot / 📞 Call Back / 💬 Quote / 📝 Add Notes. Context captured in the moment, not 2 weeks later.
📊
Dashboard Real-time web dashboard — filter, search, export. Team analytics, event breakdown. Managers see activity live, not after the fact.
🏆
Compete Live leaderboard in the team group. Rahul: 28 | Amit: 22. Gamification turns a chore into a competition. Teams capture 2-3x more.

Why I Built It This Way

These decisions are what separate a product from a toy.

Why Telegram, not a native app?

At a busy booth, you have 5-10 seconds between visitors. Ask them to download an app → create account → login → figure out the UI? Dead on arrival. Telegram is already on their phone. Open chat, snap photo, done. Zero onboarding friction.

Why GPT-4o Vision, not traditional OCR?

Indian business cards are chaos — Hindi mixed with English, creative layouts, gradient backgrounds, embossed metallic text. Tesseract and traditional OCR fail on 30%+ of them. GPT-4o handles all of it natively, including extracting semantic context (industry, services) that rule-based OCR can't.

Why multi-tenant from Day 1?

Most builders would ship single-user, then 'add teams later.' That's a rewrite, not a feature. I built workspace isolation, team roles (owner/admin/member), RBAC, and per-workspace events into the foundation. The architecture scales without rearchitecting.

Why leaderboards matter more than features?

Day 2 of a 5-day expo. Energy drops. Team is on their phones. Then a message hits the group: 'Rahul: 28 🔥 | Amit: 22 | Priya: 19.' Nobody wants to be last. Gamification without building gamification — just a sorted list and human nature.

Three Products. One Platform.

🤖

Telegram Bot

The capture layer. Multi-event, team roles, rapid-fire mode, back-side card merge, duplicate alerts, leaderboard. Zero install.

16 modules 303 tests

FastAPI Backend

JWT auth, workspace isolation, RBAC. 7 API routers — auth, workspaces, events, leads, members, analytics, invites. PostgreSQL with Alembic migrations.

7 routers REST API
📊

Next.js Dashboard

The management layer. Leads table with filters + search, team management, event analytics, CSV export, workspace settings. Responsive, real-time.

Next.js 14 Tailwind
✨ The Serve

What Got Shipped

<30s Card to CRM vs 2-3 weeks manual
95%+ OCR Accuracy Including Hindi/Gujarati
$5/mo Running Cost Server + API
10 days Idea to Production Bot + API + Dashboard
303 Tests Passing CI-ready
14 Fields Extracted Per card scan
Workspaces Multi-tenant from Day 1
0 App Downloads Telegram-native

Why Existing Tools Don't Cut It

Feature LeadSnap CamCard HubSpot
AI OCR (GPT-4o) ❌ Old OCR ❌ Basic
Zero install (Telegram) ❌ App ❌ App
Team + events ✅ HubSpot only
Hindi/regional cards Poor Poor
Real-time leaderboard
Multi-tenant
Per-event pricing ₹999 $50/yr/user $50/mo/user
🚀 The Vision

Not a Card Scanner. A Category.

LeadSnap started as "scan cards faster." But the real opportunity is owning the entire post-event pipeline — from card to closed deal.

✅ Shipped

Phase 1: Capture

  • • Telegram bot with GPT-4o OCR
  • • Team management + events
  • • Duplicate detection
  • • Leaderboard + gamification
  • • Web dashboard + API
🔄 Building

Phase 2: Platform

  • • Self-serve onboarding
  • • WhatsApp bot
  • • CRM integrations (Zapier, HubSpot)
  • • Badge / QR code scanning
  • • Billing + plans
🎯 Next

Phase 3: Intelligence

  • • Auto-research (LinkedIn, company data)
  • • Fitment scoring against ICP
  • • AI follow-up sequences
  • • Content matching engine
  • • Event ROI analytics

The insight: Every exhibition booth in India — from a 10-person startup at a local trade show to a Fortune 500 at Bharat Mandapam — has the same broken workflow. Card → bag → Excel → generic email → dead lead. The tech to fix it (GPT-4o, messaging APIs, real-time dashboards) exists today. Nobody had stitched it together for the Indian market with zero friction.

LeadSnap isn't a card scanner. It's the missing data layer between "met someone at an event" and "closed the deal."

Tech Stack
PythonFastAPIPostgreSQLNext.jsTailwind CSSGPT-4o VisionTelegram Bot APIJWT AuthAlembichttpxPydantic
Your Turn

This is what I do.

I find problems that cost companies real money. I design systems that solve them. And I ship in days, not months. If you've got something that needs building — let's talk.