RapidMind AI

RapidMind AI is an AI Agent for automated complaint resolution and transaction reversal in fintech. It runs efficiently on low-resource infrastructure - no GPUs or heavy cloud dependencies.

Comment

Inspiration

In 2023, I stopped using a popular Nigerian bank’s app after repeated transaction failures. Each failed transfer meant a stressful trip to the bank and a long wait for reversals — often over ₦2,000 lost entirely. Even when I switched to digital wallets, the same problems persisted, just with slightly faster support. As a backend intern at a fintech company, I saw firsthand how even internal teams struggled with these issues. I later experienced a delayed ₦25,000 reversal that took three days to resolve. These aren’t isolated events - across Africa, failed transactions are rising, leaving users frustrated and unsupported. In 2023, I wrote a LinkedIn post about using AI to solve this. Now, in 2025, we’ve built that idea into reality.

What it does

RapidMind AI is an always-on virtual agent that automatically resolves failed fintech transactions and initiates reversals using natural language processing (NLP). It mimics a human support agent by:

  • Understanding complaints: It parses customer messages to identify issues (e.g. failed airtime top-ups, failed transfer, or a debit without credit).
  • Verifying details: It checks transaction records and logs to confirm what happened.
  • Providing updates: It communicates real-time resolution timelines to users. By operating 24/7, RapidMind AI reduces user stress and restores trust when transaction issues arise.

How we built it

User research: We started with a survey to gather firsthand feedback on common transaction failures and user pain points. Design & planning: We analyzed the survey data and brainstormed solution designs to address these issues. NLP engine: After research, we selected Rasa NLU for its robust intent classification and fallback handling capabilities. Tech stack: We used modern frameworks and tools for each component:

  • Next.js for the frontend user interface
  • Node.js for the backend server and API logic
  • PostgreSQL for storing transaction and user data
  • Redis as an in-memory cache for fast data access
  • Socket.IO to handle real-time event updates between client and server
  • Redux for state management in the frontend application
  • Whimsical for creating architectural diagrams and design flows

Implementation: We built a conversational workflow where RapidMind AI recognizes user intents (for example, “report failed transaction”) and runs custom actions in Rasa to extract key details (such as transaction IDs and user account information). The system then attempts to resolve the issue by querying transaction logs or calling reversal APIs, and it informs the user of the outcome.

Challenges we ran into

A major challenge was intent classification and extracting key details like transaction IDs. We initially tried Hugging Face’s zero-shot classifier but found it insufficient for our needs, especially for handling custom actions. After further research, we switched to Rasa NLU, which offered better accuracy, support for custom actions, and clearer documentation that helped us reach MVP. We also faced issues with real-time event handling, which we resolved using Socket.IO for smooth client-server communication.

Accomplishments that we're proud of

We successfully implemented all planned features and user flows in line with the hackathon requirements. A fully functional prototype of RapidMind AI is currently running locally, demonstrating its core capabilities. Key accomplishments include robust intent recognition, custom actions for extracting transaction details, and a complete flow for detecting and resolving failed transaction scenarios end-to-end.

What we learned

We learned that users facing issues, especially with their money, want fast and stress-free solutions. Building with the right tools and the right team makes even complex challenges achievable. Most importantly, innovation doesn't always mean reinventing the wheel. Reusing efficient, well-supported tools can help solve real-world problems faster and more effectively.

What's next for RapidMind AI

Next, we plan to deploy our solution to the cloud and begin user testing with real transaction data (where permitted). We’ll continue refining the NLP model and improving real-time response capabilities. Beyond the tech, we aim to engage with fintech startups and traditional banks to explore integration opportunities and potential partnerships that can bring this solution to market.

Built With

Share this project:

Updates