BrandPilot — AI Marketing Co-Pilot

landing page
backend page
loading screen
report mail
sns
company detals
dashboard
timelines actions
strategy
connectors
insights

The Problem

400 million small businesses, indie creators, freelancers, and studios worldwide have no marketing team. A real marketing agency costs $3,000–$10,000/month. Most SMBs simply cannot afford it — so they spend 15–20 hours every week manually checking competitor prices, reading reviews, searching for opportunities, and trying to keep up with market trends.

BrandPilot fixes that.

What It Does

BrandPilot is an AI-powered marketing co-pilot that runs a fleet of 4 parallel Amazon Nova Act browser agents overnight on AWS — automatically doing everything a marketing agency would do, delivered to your inbox every Monday morning.

The 4 Nova Act Agents:

🕵️ Competitor Scout — Browses competitor websites, extracts pricing, promotions, and recent changes
⭐ Review Guardian — Monitors Google, Yelp, and Trustpilot for new customer reviews
📣 Mention Tracker — Scans Reddit, news sites, and blogs for brand mentions and trending topics
💰 Opportunity Hunter— Finds grants, sponsorships, and brand deals matching your business profile

Amazon Nova 2 Lite then synthesizes all agent outputs into a prioritized weekly intelligence digest — competitor gaps, review replies needed, opportunities with deadlines, and a ranked action plan — delivered via Amazon SNS.

What Inspired Me

I kept seeing the same problem everywhere. A small café owner spending Sunday nights manually checking what Starbucks is doing. A freelancer trying to compete with Zoho without knowing their pricing strategy. A solo creator missing grant deadlines because nobody told them the opportunity existed.

These businesses don't need a $10K agency. They need one intelligent agent working for them 24/7. Amazon Nova Act made that possible.

How I Built It

Tech Stack:

Amazon Nova Act— Browser automation agents using AgentCore Runtime
Amazon Nova 2 Lite — Intelligence synthesis via Amazon Bedrock
AWS Lambda — Serverless agent execution (containerized)
Amazon S3 — Persistent storage for runs, strategy, timeline, digests
Amazon SNS — Weekly digest email delivery
Amazon EventBridge— Scheduled Monday morning agent runs
FastAPI + React + TypeScript — Full-stack web application
AWS SAM — Infrastructure as code deployment

Architecture The frontend is a React SaaS dashboard deployed on S3. The backend is a FastAPI service running on Lambda via API Gateway. When a user clicks "Run AI Scan", AWS Lambda triggers all 4 Nova Act agents simultaneously using Python ThreadPoolExecutor — each agent navigates real websites using natural language act() commands and extracts structured data via Pydantic schemas. Results are stored in S3, synthesized by Nova 2 Lite via Bedrock, and delivered via SNS.

Challenges I Faced

Nova Act in Lambda Running Playwright-based browser automation inside AWS Lambda containers was the hardest technical challenge. Lambda's ephemeral environment required custom Chromium setup, specific browser flags, and switching to AgentCore Browser backend to run browsers reliably at cloud scale.
Parallel Agent Reliability Running 4 agents simultaneously without timeouts or resource conflicts required careful concurrency management — limiting live mode workers and implementing per-task fallback handling so the dashboard always shows useful output even when individual agents hit guardrails.
India Region Access Nova Act's web playground is restricted in India. The entire development workflow had to be built through the Python SDK and AWS CLI — which actually pushed me to build a more production-grade architecture than I would have otherwise.
Making It Non-Technical The hardest UX challenge was designing for a café owner or freelancer who has never used AWS. The 3-step onboarding wizard, runtime transparency badges, and plain-English digest format were all designed to make enterprise-grade AI accessible to anyone.

What I Learned

Amazon Nova Act is genuinely powerful for real-world browser automation — not just demos
Building production multi-agent systems requires serious thought about failure modes, fallbacks, and user trust
The best AI products hide the complexity and show only the value
A small business owner doesn't care about AgentCore — they care that their Monday morning email tells them Starbucks just dropped prices