Fixora : Customer Support Multi-Agent System

Inspiration

Fixora AI was inspired by our own frustrating experiences with warranty claims and electronic device repairs. We've spent hours going back and forth to computer shops, only to discover that replacement parts weren't available or that another visit was necessary.

Even getting an estimated repair cost or turnaround time often required multiple phone calls or physical visits. We wanted to solve a simple problem: What if customers could check warranty eligibility, spare-part availability, and estimated repair costs without visiting the shop?

That's how Fixora AI was born.

What it does

Fixora AI is a multi-agent AI platform that simplifies warranty claims and repair management for electronics shops.

Customers upload their invoice, product images, and a description of the issue. Specialized agents then work together to:

Extract purchase and warranty information from invoices.
Analyze product images for visible damage.
Verify warranty eligibility against shop policies.
Identify compatible replacement parts and check inventory.
Generate evidence-backed repair assessments and cost estimates.

When evidence is missing or uncertain, the system escalates the case for human verification rather than making unreliable decisions.

How we built it

We built Fixora AI using Jac and byLLM, combining agentic AI with deterministic business rules.

The system coordinates four specialized stages: Evidence Analysis, Warranty Coverage, Repair & Inventory, and Resolution.

Jac's graph-based architecture manages relationships between cases, evidence, policies, and repair options. We use MongoDB and GridFS for persistent case data and uploaded files.

AI interprets unstructured information, while policy agent handles warranty calculations, inventory compatibility, and pricing.

Challenges we ran into

Our biggest challenge was ensuring that AI-generated assessments were reliable enough for real-world warranty decisions.

Images cannot always reveal the actual cause of hardware failures, and warranty policies require precise validation. We addressed this by separating AI interpretation from deterministic decisions and introducing human verification for uncertain cases.

Coordinating multiple stages, maintaining consistent data, and integrating everything within a limited hackathon timeframe were additional challenges.

Accomplishments that we're proud of

Building a coordinated multi-agent workflow using Jac and byLLM.
Integrating invoice analysis, damage assessment, warranty verification, and inventory matching into one platform.
Designing evidence-backed decisions that prioritize reliability over AI-generated guesses.
Implementing persistent case management with MongoDB and GridFS.
Creating a practical solution inspired by a problem we've personally experienced.

What we learned

We learned that building trustworthy agentic AI isn't about automating every decision, but knowing when AI should make an interpretation and when deterministic rules or human judgment should take over.

We also gained practical experience with Jac's graph-based programming model, byLLM integration, multi-agent orchestration, and designing workflows that handle uncertainty.

What's next for Fixora : Customer Support Multiagent System

We envision expanding Fixora AI into a complete customer support and after-sales service platform, and broaden the support for the capabilities in customer support in terms of sophisticated issue resolving while providing support for different products categories.

Our future plans include WhatsApp-based customer interactions, dedicated customer and technician portals, real-time inventory and supplier integrations, predictive repair turnaround estimates, and support for more electronic devices.

Our goal is to make warranty claims and repairs faster, more transparent, and more convenient for both customers and electronics shops.