PromoPilot

AI that markets while you build

Comment

Inspiration

At my previous startup, marketing and DevRel were handled by just one or two people doing the work of five. They had to learn the SDK, prepare demos, field every kind of hacker question, post on social media, convince developers to try the SDK, and craft prizes and offers. I kept thinking: what if a platform could take product info once and figure out multiple ways to promote it automatically?

What it does

PromoPilot turns a short product brief into a full multi-channel campaign that improves itself over time.

  • Intake once : Paste product info, goals, audience, tone, target platforms, posting window, and asset preferences.
  • Generate campaign assets : Creates post copy variations, CTA suggestions, hashtags, captions, alt text, and thumbnail prompts aligned to goals.
  • Create media on demand : Produces voiceovers and short videos from prompts and scripts. Returns playable links and stores metadata for reuse.
  • Plan and schedule : Builds a posting calendar across channels with recommended days and times. Lets you approve, edit, or auto-schedule.
  • Publish and fetch : Posts to social directly, attaches media, and can fetch post details by URL or ID for tracking.
  • Iterate for lift : Runs improvement loops that try new hooks, formats, and lengths based on impressions, engagement, and early feedback.
  • Review : Provides one-click approvals, edit suggestions, and quick regenerations for any asset.
  • Compliance and safety checks : Screens for banned terms, missing disclosures, and platform guideline issues before publishing.
  • Metrics at a glance: Tracks reach, clicks, watch time, and CTR per post and per channel, and rolls up to campaign KPIs.

Example flow
Brief in, assets out. You enter a product blurb and audience. PromoPilot drafts posts and media prompts, generates an explainer clip with voiceover, schedules a Twitter and LinkedIn launch, publishes, monitors performance, and iterates with stronger hooks until KPIs are met.

How we built it

Architecture Diagram

Architecture Overview

  • Client
    Sends product details, offers and target audience to the backend and receives media, post conetent and a promotion plan form the agent.

  • Backend (MCP Server) — AWS Bedrock
    Central hub. Stores and reads data from AWS RDS, calls Agent 1 for content, calls Agent 2 for media, coordinates with Agent 3 for posting, and loops with Agent 4 for improvements.

  • AWS RDS
    Persists campaigns, assets, media job metadata, and publish events.

  • Agent 1: Content Generation
    Generates campaign text assets from the brief and returns them to the backend.

  • Agent 2: AWS + Fetch.ai (Mailbox uAgent, ChatProtocol enabled) — Available on AgentVerse Marketplace
    Media generation pipeline using AWS Polly (Audio) and AWS Nova Reel (Video). Returns task IDs and media links to the backend. Link: https://agentverse.ai/agents/details/agent1q0xw02drl892dtx6345ts5d3rmrefw6gl7a8l29rwqwm3wahefvqugazm8w/profile

  • Agent 3: Social Media Post Scheduling + Triggering (Mailbox uAgent, ChatProtocol enabled) — Available on AgentVerse Marketplace
    Posts text and media and can fetch post details. Supports uAgent ↔ uAgent communication with Agent 2 for media handoff. Link: https://agentverse.ai/agents/details/agent1qthq0qvvq302ufxnsxx8ja4srm2q7npzrkvjm0pssas5t8d00nffv6v7udh/profile

  • Agent 4: Reiteration and Content Improvement Agent
    Feeds performance signals back to the backend and returns improved content for the next iteration.

Data flow note: All traffic is brokered by the backend; the only direct agent-to-agent path is the uAgent–uAgent communication between Agent 2 and Agent 3 for media posting. Donw using Fetch.Ai

  • Deployment
    The two uAgents are public on Agentverse so other apps can reuse them. ChatProtocol is also enabled on both of them. The web app, MCP server, and utilities are deployed behind HTTPS with CORS configured for the frontend. Link to the uagents: twitter-ops-agent, aws-bedrock-media generator-agent

Sponsor Prizes Targeted

  • Best use of Claude AI:
  • FetchAI: Best use of FetchAI (Deployed two uagents on agentverse, enabled chat protocol)
  • FetchAI: Best use of Best Deployment of Agentverse (Deployed two uagents on agentverse, enabled chat protocol)
  • FetchAI: Most Viral ASI:One Personal AI (Used ASI using chat protocol to write an engaging tweet about FetchAI's agentverse and ASI, Tweet Link: https://x.com/OrdinaryLunati2/status/1982289300600549533)

Challenges we ran into

  • Orchestrating multiple agents into one smooth system. We wired together four agents for marketing: post generation with Claude, media generation with AWS and FetchAI, social scheduling and triggering on Agentverse, and a reiteration and content improvement loop with uAgents and Claude.
  • Social media integration. Twitter’s free tier rate limits made integration and testing tricky, especially wrapping logic inside a uAgent cleanly.
  • Package dependency tangles. With Flask, boto3, Chakra UI, uAgents, and more, version conflicts cost time. The Fetch.ai team helped me resolve uagent_core issues, and disciplined debugging got the stack stable.

Accomplishments that we're proud of

  • Built a multimodal agentic system that helps small teams market effectively when budgets and headcount are tight. That feels useful and real.
  • Finished despite slow or no internet. I pushed through long nights, found better network windows, and kept going even when I wanted to go home. I am proud of the grit.
  • Proved to myself I can still ship without AI at my fingertips. I debugged third-party SDK issues with docs and prints alone, which rebuilt confidence.

What we learned

  • How to assemble a mini-modular multi-agent system, stitch it to an MCP server, and make the UX feel like magic.
  • A lot of practical AI from talking with other hackers and sponsors.
  • The value of an agent marketplace. Fetch.ai’s Agentverse made feature integration faster and cleaner. I learned core ideas like uAgents, Agentverse, and ASI One.
  • How to integrate AWS Bedrock for better cost and token control, and approaches that improve longer video generation.

What's next for PromoPilot

  • Add payments using the Stripe Payment Agent from Fetch.ai to offer premium subscriptions.
  • Expand platform integrations: LinkedIn, Instagram, Facebook Marketplace, Reddit, and more, with a dedicated uAgent for each so others can reuse them by address.
  • Make the AI more decisive. Automatically double down on channels and formats that perform better, and deepen competitive scraping for smarter moves.

Built With

Share this project:

Updates