Inspiration
AI agents are becoming capable of shopping and making payments on our behalf. But can we really trust them with our money? I wanted to solve this problem by building a layer that gives users control over what their AI agents can spend and do.
What it does
Veltrio is a financial trust layer for AI agents.
- Create spending mandates using natural language.
- Set budgets, merchant restrictions, and expiry dates.
- Discover products through AI-powered shopping workflows.
- Evaluate every purchase with ALLOW, BLOCK, or ESCALATE decisions.
- Explore payment workflows through PayPal Sandbox.
- Track spending and financial activity.
It helps users enjoy autonomous shopping without giving agents unlimited financial authority.
How we built it
I independently designed and developed Veltrio using Next.js, React, TypeScript, Tailwind CSS, and PostgreSQL.
I integrated Google Gemini for natural-language mandate extraction, Channel3 for product discovery, PayPal Sandbox for payment workflows, and Render for deployment.
The most important architectural decision was separating AI interpretation from deterministic payment authorization. AI can understand the user's instructions, but the authorization gateway enforces the actual rules.
Challenges we ran into
Since I built the entire project alone, managing frontend, backend, AI integrations, databases, and payment workflows within the hackathon timeline was challenging.
I faced API failures, Gemini model availability issues, response timeouts, and integration difficulties. I also had to prioritise essential features over several ideas I wanted to implement, including voice shopping and messaging-platform integrations.
Late-night debugging, researching documentation, and repeatedly testing different approaches became part of the journey.
Accomplishments that we're proud of
- Built the complete Veltrio application independently.
- Developed a natural-language spending mandate workflow.
- Designed a rule-based authorization gateway with three clear decisions.
- Integrated product discovery and PayPal Sandbox workflows.
- Brought the application online using Render.
- Turned an idea about AI payment safety into a working product prototype.
What we learned
I gained practical experience in full-stack development, AI integration, payment APIs, database design, and deployment.
More importantly, I learned that building trustworthy AI systems requires more than intelligence. It requires clear boundaries, predictable decisions, and user control.
What's next for Veltrio
I want to take Veltrio beyond a hackathon prototype by:
- Supporting more AI agents and shopping platforms.
- Expanding payment integrations and transaction verification.
- Introducing voice-based and messaging-based shopping.
- Improving fraud detection and spending analytics.
- Building stronger authentication, security, and user approval workflows.
My long-term vision is to make Veltrio the trust and authorization infrastructure for the growing agentic commerce ecosystem.
Built With
- agents
- ai
- api
- channel3
- css
- e-commerce
- fintech
- gemini
- natural-language-processing
- next.js
- paypal
- postgresql
- react
- render
- tailwind
- typescript