TunjiaX Wallet

TunjiaX: Eliminating banking barriers with Gemini-powered intelligence, Eleven-Lab's voice tech and secure biometric verification for seamless, hands-free Transfers.

Comment

Here is the full text formatted in Markdown.

TunjiaX: Conversational Finance for Every Generation

🌟 The Inspiration: Breaking the Digital Divide

In Nigeria, a significant digital divide exists. For our elders—those aged 60 and above—the rapid shift to digital banking has been more of a barrier than a blessing. Many did not grow up with touchscreens and complex UI layouts. I often witnessed seniors calling on strangers to help them navigate their banking apps—a desperate act of trust that frequently leads to compromised security and devastating scams.

On the other end of the spectrum, the "instant" generation demands speed. For them, typing is a chore; they want to multitask, speak, and move on.

TunjiaX was born from these two extremes. I wanted to build an assistant that was human enough for my parents to trust and fast enough for my peers to use. Thus, Nugar was created: a voice-first AI agent that makes banking as simple as having a conversation.

🛠️ How I Built It

TunjiaX is powered by a high-performance stack designed for low latency and high security:

  • The Brain: Gemini 2.0 Flash (Vertex AI) handles the complex task of extracting NUBAN details, bank names, and transaction intent from natural speech.
  • The Voice: ElevenLabs provides the lifelike vocal interface, giving Nugar a personality that feels helpful rather than robotic.
  • The Security: A multimodal biometric system. We don't just rely on voice; we use Gemini’s vision capabilities to perform liveness checks and face verification.
  • The Backend: A custom FastAPI server hosted on Google Cloud Platform (GCP).

To ensure the agent felt responsive, I implemented an asynchronous streaming architecture. The efficiency of the interaction can be represented as:

Where represents the time taken to complete a transaction. By moving to a voice-first UI, we reduce the cognitive load and time-to-completion significantly.

🏗️ The Custom OpenAI Middleware

Because I joined the hackathon late and hackathon-specific ElevenLabs credits were exhausted, I had to innovate. I built a custom OpenAI-compatible REST layer from scratch. This allowed me to map Gemini’s unique responses into the standard chat/completions format required by ElevenLabs' Custom LLM integration. This "wrapper" logic ensured that Nugar could speak and trigger tools (like biometrics) simultaneously without hitting the limits of the standard free tier.

⚡ Challenges & Resilience

Building TunjiaX was a race against time and infrastructure:

  • The Financial Hurdle: I spent two full days—40% of my remaining time—simply trying to find a Nigerian debit card that Google Cloud would accept for billing. I finally succeeded only 72 hours before the deadline.
  • The Credit Crunch: With ElevenLabs credits running low, I had to optimize every single API call, leading to a much cleaner and more efficient prompt engineering strategy.
  • The 5-Day Sprint: I discovered the challenge late. While others had weeks, I had 120 hours. I had to quickly set up the GCP environment, write the banking tools, and polish the UI while battling a ticking clock.

📚 What I Learned

This project taught me that accessibility is security. When we make tools easier for the elderly to use themselves, we remove the need for them to share their pins and passwords with "kind strangers."

I also deepened my knowledge of Large Language Model (LLM) orchestration—learning how to "close the loop" between function calls and conversational responses. TunjiaX isn't just a wallet; it’s a proof of concept that the future of the Nigerian economy is vocal.

I am proud to have met the deadline, proving that with the right tools (and a card that finally works!), anything is possible.

Built With

Share this project:

Updates