ClaimNova

An Active Agent that completely automates e-commerce returns, cross-referencing user photos with PostgreSQL data to prevent fraud.

Comment

💡 Inspiration

Return fraud and manual damage assessments cost the e-commerce industry billions of dollars annually. Traditional chatbots fail here because they are stateless and cannot securely interact with backend systems. We wanted to build a true Active Agent—an autonomous workforce that could actually see physical damage, securely query a database to verify a user's purchase history, and execute actions across multiple platforms without human intervention.

⚙️ What it does

ClaimNova is a fully autonomous, multi-system claims adjuster orchestrated entirely by Airia.

  1. Intake & Verification: The Airia agent greets the customer and uses native tool calling to autonomously query the company's PostgreSQL database to verify the purchase.
  2. Multimodal Evidence Collection: It asks the user to upload a photo of the damaged item.
  3. Cross-System Fraud Detection: ClaimNova compares the user's photo against the database text record. If the database states the user bought a "Dell Monitor," but they upload a photo of a shattered "Lenovo Laptop," it instantly detects the fraud and rejects the claim.
  4. Action Execution: If valid, the Airia agent dynamically generates a safe SQL query to update the database state, and simultaneously triggers a webhook to notify the admin team via Slack.

🛠️ How we built it

Airia serves as the master orchestrator for this entire architecture, acting as the brain stem between the user interface, the reasoning model, and the backend database.

  • The Orchestrator: We used Airia's powerful node-based workflow to manage the conversational state and execute Model Context Protocol (MCP) tools.
  • The Brain: We plugged Amazon Nova into Airia to handle the multimodal vision analysis and strict JSON reasoning.
  • The Database (Tool Calling): Airia executes native SQL queries directly against our Supabase PostgreSQL database, reading user records and writing approval statuses in real-time.
  • Human-in-the-Loop (HITL): Validated claims are pushed to a live Admin Dashboard and Slack channel where a human makes the final financial decision, showcasing a perfect autonomous-to-human handoff.

🚧 Challenges we ran into

Orchestrating an LLM to directly execute PostgreSQL queries is inherently risky. Our biggest challenge was engineering the Airia agent to cleanly separate its conversational responses from its background tool executions. By relying on Airia's native tool-calling loops, we were able to instruct the agent to silently fetch database rows, inject them into its context window, and then reply to the user, creating a seamless, lag-free frontend experience.

🏆 Accomplishments that we're proud of

We successfully achieved true multi-system orchestration. Watching the Airia agent autonomously pause a chat, execute a SELECT query in Supabase, analyze an image, execute an UPDATE query, and fire a webhook to Slack—all within seconds—proves that Active Agents are the future of enterprise automation.

📚 What we learned

We learned that the true power of an AI model is unlocked entirely by its orchestrator. By utilizing Airia's framework, we transformed a standard conversational LLM into an intelligent backend administrator capable of securely managing enterprise data states.

Built With

Share this project:

Updates