Input
Output

BizBoost AI — Project Story

Inspiration

Walking through any Indian bazaar, you see incredible products — handmade jewellery, fresh street food, tailored clothes. But ask the vendor if they're on WhatsApp Business or Instagram, and most will shake their heads. Not because they don't want to grow, but because creating professional digital content requires English fluency, design skills, and time — three things most small business owners simply don't have.

India has 60 million small businesses. Only 15% have any digital presence. 85% rely entirely on walk-in customers.

We asked ourselves: what if all they had to do was speak?

That question became BizBoost AI — बोल के बेचो. Speak it. Sell it.

What it does

BizBoost AI lets any small business owner create a professional, bilingual WhatsApp catalog post in seconds — just by speaking in Hindi.

Speak — Tap the mic and describe your product in Hindi ("पनीर टिक्का, बहुत टेस्टी, सिर्फ 200 रुपीस")
Generate — Amazon Bedrock's Nova Lite model creates a polished, emoji-rich catalog post in both Hindi and English
Share — Copy the post directly to WhatsApp Business with one tap

No English needed. No design skills. No expensive tools. Just speak and sell.

Every generated post is automatically saved to Amazon DynamoDB, giving business owners a growing catalog history.

How we built it

We built BizBoost AI as a fully serverless, 100% AWS-native stack — deployed end to end in 48 hours.

Frontend:

Next.js with Framer Motion for smooth animations
Browser Web Speech API for Hindi voice capture (hi-IN)
Hosted on AWS Amplify with automatic GitHub deployments

Backend:

Pure AWS Lambda function (Python) — no servers, no containers
Amazon API Gateway exposing the /generate endpoint
Amazon Bedrock (Nova Lite) for bilingual content generation
Amazon DynamoDB for storing all generated posts
Deployed using AWS SAM (Serverless Application Model)

AI Prompt Engineering: We crafted a structured prompt that instructs Nova to understand Indian product descriptions, detect prices mentioned in Hindi speech, and output formatted WhatsApp-ready posts with emojis in both Hindi and English.

Challenges we ran into

1. Zappa vs SAM We initially tried Zappa to deploy our Flask backend to Lambda. It took hours of permission errors and IAM debugging before we made the call to switch to AWS SAM — which deployed cleanly in minutes. Lesson learned: use the tool built for the job.

2. Mangum compatibility After switching to SAM, we tried using Flask + Mangum as the Lambda handler. Mangum kept throwing KeyError: 'sourceIp' due to API Gateway v1 vs v2 format mismatches. We ultimately ditched Flask entirely and rewrote app.py as a pure native Lambda handler — which was actually simpler, faster, and had zero dependencies.

3. CORS on API Gateway Getting CORS headers to pass through correctly from Lambda to the browser took significant trial and error. The fix required explicitly configuring GatewayResponses in the SAM template to add CORS headers even on 4XX and 5XX error responses.

4. Python 3.13 vs Lambda runtime Our local environment used Python 3.13 but Lambda only supports up to 3.12. Building with sam build --use-container and Docker resolved this by compiling in the exact Lambda Linux environment.

Accomplishments that we're proud of

🚀 Built and deployed a fully working product in 48 hours — live at a real AWS Amplify URL
🎤 Real Hindi voice recognition — not a simulation, actual browser speech API capturing live Hindi speech
🤖 Amazon Nova integration — first time either of us used Amazon Bedrock, got it generating quality bilingual content
☁️ 100% serverless AWS stack — Lambda, API Gateway, DynamoDB, Amplify — no EC2, no servers, no maintenance
📱 Actually useful for real people — tested with real Hindi product descriptions, outputs are WhatsApp-ready

What we learned

AWS SAM is the right tool for serverless — it handles Lambda, API Gateway, IAM, and CloudFormation in one clean template
Flask isn't necessary for Lambda — a native Python Lambda handler is simpler, faster, and has zero cold-start overhead from extra libraries
Prompt engineering matters — the quality of Nova's output changed dramatically based on how we structured the prompt, especially for bilingual Hindi/English output
Ship first, polish later — getting a working demo deployed mattered more than perfect code. We iterated on the live URL rather than local
Two-person teams move fast — with clear ownership (frontend vs backend) and AI-assisted development tools, a small team can outpace larger ones

What's next for BizBoost AI

Phase 2 (Next 3 months):

🌍 Multi-language support — Tamil, Telugu, Bengali, Marathi (Amazon Transcribe for 10+ Indian languages)
📸 Smart Photo Enhancer — Amazon Rekognition to auto-describe product photos and generate posts from images
📲 Direct publishing — One-tap post to WhatsApp Business, Instagram, and Facebook via their APIs
🏪 Business profiles — Shop name, category, and location personalize every Nova-generated post

Phase 3 (6 months):

🎉 Festival content scheduler — Auto-generate Diwali, Eid, Holi promotional posts ahead of time
💬 AI customer bot — Responds to common buyer queries (price, availability, delivery) in the customer's language, 24/7
📊 Analytics dashboard — Simple metrics: views, best performing posts, peak times
💳 Monetization — Free tier (5 posts/day) + Pro (₹199/month) + Business (₹499/month)

BizBoost AI isn't just a hackathon project — it's a genuine solution for the 60 million Indian small businesses still waiting to go digital. बोल के बेचो. Speak it. Sell it. 🚀