Rwadaocom

For #MemoryAgent: AI learns from on-chain sales (not likes). Each sale+referral = signal. Agent learns drivers. Video=camera; memory re-ranks in real-time. Knowledge to RWA-DAO. Demo: shift after sale

Comment

MemoryAgent: For the hackathon, we're building an AI Influencer Studio that learns from real revenue, not fake likes.

Our RWA-DAO/XDC toolkit turns every on-chain sale (dollar amount + creator referral) into an un-fakeable signal. While others optimize vanity metrics, our Memory Agent learns creator-by-creator what actually drives purchases: platform, style, timing.

The video model is just the camera. The memory is the creative director that improves with every sale—re-ranking next content batches in real-time. Human-in-the-loop ensures brand safety.

Proprietary value: the "what converts" knowledge belongs to RWA-DAO and appreciates with use. Future: views, clicks, shares, geo data. Demo shows live preference shift after simulated sale.

Inspiration

Create an easy to use, agentic ai influencer studio toolkit for our rwa-dao / xdc community to easily generate content for posting to their social media accounts that is learning what works and improving over time

What it does

Allows users to generate content using their own images or short video to generate content using ageitic ai to create new content for the rwadao community on social media, then monitors and improves the next batch of content

How we built it

We used various LLMs to generate short video using ai from various user inputs and rwa images of the luxury watches by rwa dao

Challenges we ran into

Token limitations , we ran into limitation of tokens being used to generate videos

Accomplishments that we're proud of

By decentralizing the effort to generate videos and providing the raw content for users to select to generate their own videos we have manage to overcome the limitations set by overcoming the challenge as a community

What we learned

Unity is strength

What's next for Rwadaocom

More enhancements for the community

Built With

  • alibabacloud
  • claude
  • deepseek
  • grok
  • happyhorse
  • higgsfield
  • nanobanana
  • qwen
  • stablediffusion
Share this project:

Updates

posted an update —

The "Browser demo: https://rwadao.netlify.app/" is the RWA-DAO Creator Studio DEMO for the Qwen hackathon.

The https://rwa-web-gold.vercel.app/ deployment is the simulated NFT minting application developed earlier on for RWA-DAO and modified for simulation for this DEMO as a reference only.

Testing Instructions and use: Browser demo: https://rwadao.netlify.app/ Alibaba Cloud deployment: Function Compute in Singapore (ap-southeast-1), function Rwadao-qwen-ai-hackaton. Public endpoint: https://rwadao-hackaton-ardhtlwfjj.ap-southeast-1.fcapp.run The Alibaba endpoint returns HTTP 200. Its shared fcapp.run domain may force HTML downloads, so use the Netlify URL for the Creator Studio UI demo. No credentials are required.

The Creator Studio is the content-generation frontend; the learning loop is demonstrated by the dedicated MemoryAgent judge lab included in the hackathon build. Judge test scenario: Start with Reset memory. The agent displays: “No performance history yet — the agent is exploring.” Select the outcome “Brazil · TikTok · Luxury Reveal drove $800 in mints.” Observe that: the reinforced-outcome counter increases; the TikTok, Brazil and Luxury Reveal memory weights increase; the next content batch is re-ranked; the exact Qwen prompt changes to include the newly learned preference.

Select “Argentina · Instagram · Unboxing ASMR drove $1,500 in mints.” The memory, rankings and Qwen prompt adapt again toward the stronger revenue signal. Press Reset memory to return to the original baseline. This demonstrates the complete loop: Simulated on-chain mint outcome → persistent creator memory → learned weights → re-ranked content → updated Qwen prompt The interactive lab uses deterministic simulated mint outcomes, with no API key or cost, but it runs the same tested MemoryAgent functions as the hosted scheduler. In production, the reinforcement signal is designed to come from indexed XDC Minted events. Judge-lab source: https://github.com/chris-blvck/rwa.dao/blob/main/frontend/components/rwa/screens/HackathonScreen.tsx Memory core: https://github.com/chris-blvck/rwa.dao/blob/main/frontend/lib/rwa/agent_memory.ts Alibaba judge-lab route: https://rwadao-hackaton-ardhtlwfjj.ap-southeast-1.fcapp.run/hackathon Note: Alibaba’s shared fcapp.run domain may download the HTML instead of displaying it. The same interactive scenario is shown clearly in the public demo video.

Log in or sign up for Devpost to join the conversation.

posted an update —

The Creator Studio is the content-generation frontend; the learning loop is demonstrated by the dedicated MemoryAgent judge lab included in the hackathon build. Judge test scenario: Start with Reset memory. The agent displays: “No performance history yet — the agent is exploring.” Select the outcome “Brazil · TikTok · Luxury Reveal drove $800 in mints.” Observe that:the reinforced-outcome counter increases; the TikTok, Brazil and Luxury Reveal memory weights increase; the next content batch is re-ranked; the exact Qwen prompt changes to include the newly learned preference.

Select “Argentina · Instagram · Unboxing ASMR drove $1,500 in mints.” The memory, rankings and Qwen prompt adapt again toward the stronger revenue signal. Press Reset memory to return to the original baseline. This demonstrates the complete loop: on-chain mint outcome → persistent creator memory → learned weights → re-ranked content → updated Qwen prompt The interactive lab uses deterministic simulated mint outcomes, with no API key or cost, but it runs the same tested MemoryAgent functions as the hosted scheduler. In production, the reinforcement signal comes from indexed XDC Minted events. Judge-lab source: https://github.com/chris-blvck/rwa.dao/blob/main/frontend/components/rwa/screens/HackathonScreen.tsx Memory core: https://github.com/chris-blvck/rwa.dao/blob/main/frontend/lib/rwa/agent_memory.ts Alibaba judge-lab route: https://rwadao-hackaton-ardhtlwfjj.ap-southeast-1.fcapp.run/hackathon Note: Alibaba’s shared fcapp.run domain may download the HTML instead of displaying it. The same interactive scenario should therefore be shown clearly in the public demo video. We should not claim that the current Netlify Creator Studio page visibly demonstrates the learning loop by itself.

Log in or sign up for Devpost to join the conversation.

Submission history