Panthoron Agentic Food Traceability

• 

  Panthωron_TraceAuditReady

• 

• 

  Accompanying_Traceability_Form

• 

  DRAFT_RECALL_REPORT

Inspiration

The concept for this project was born out of our ongoing real-world industry work. We are the creators of Panthoron-TraceAuditReady, a foundational traceability application that was recently accepted into the Exploration Phase of EIT Food (The European Institute of Innovation and Technology for Food — the European Union's leading food innovation initiative). While working with industrial food factories, we recognized a critical bottleneck: when a contaminated raw material (like E. coli) is detected, tracing it through complex ERP systems, dynamic freezing tunnels, and logistics documents takes hours of manual panic. We saw this Google Hackathon as the perfect opportunity to evolve our existing architecture. We wanted to prove how Google’s Agentic AI framework and Gemini 3.5 can elevate a traditional digital tool into a fully autonomous, decision-making ecosystem that resolves a food safety crisis in seconds. Panthoron TraceAudit acts as an autonomous digital Senior Quality Manager.

What it does

Panthoron TraceAudit is an autonomous AI agent designed to execute complex, multi-level traceability and quarantine operations instantly. Powered by Google's Gemini, it acts as a digital Senior Quality Manager. Triggered by a crisis email, the agent autonomously:Queries the ERP to trace the raw material drop time.Calculates the physical quarantine window using strict factory logic: $\text{Exit Time} = \text{Drop Time} + \text{Conveyor Time} + \text{Freezing Tunnel Time}$.Isolates the final inventory by identifying the exact Master Pallet (LPN). Scans logistics PDFs in Google Drive to intercept moving trucks, generating a structured recall report instantly.

How we built it

We evolved Panthoron from a local prototype into a live, stateless microservice deployed on Google Cloud Run. Using Python and the modern google-genai SDK, we built a robust Agentic Workflow powered by Gemini. Instead of acting as a conversational chatbot, the model is governed by strict System Instructions and a suite of custom Python tools. We exposed this intelligence via a secure API endpoint, allowing the agent to be triggered externally via simple HTTP POST requests.

Challenges we ran into

Transitioning from a conversational AI mindset to a deterministic agent was our biggest hurdle. Initially, the LLM attempted to guess times or hallucinate pallet numbers. We overcame this by heavily refining our prompt engineering and enforcing strict JSON schemas for function calling, forcing the agent to use its tools and compute the math before formulating an answer. Additionally, deploying to the cloud meant navigating complex IAM roles, billing configurations, and API rate limits.

Accomplishments that we're proud of

Overcoming the above infrastructure challenges to deliver a flawless, live "Golden Path" execution—resolving a high-stakes food safety crisis in under 20 seconds—is our proudest achievement.

What we learned

We learned that LLMs, when equipped with the right tools, are incredibly adept at solving supply chain logistics. We also learned that industrial AI requires strict guardrails; the value lies not in the AI's creativity, but in its ability to autonomously execute rigid compliance protocols faster than any human could.

What's next for Panthoron Agentic Traceability

We plan to deploy this Agentic AI Proof of Concept directly into our live B2B SaaS ecosystem. We will implement full OAuth 2.0 integrations for live Workspace OCR and connect IoT telemetry (e.g., ESP32 sensors) to read real-time machine states, piloting this breakthrough with our EIT Food manufacturing partners.

Built With

• agentic-ai
• cloud-run
• elevenlabs
• food-safety
• function-calling
• gemini-3.5-flash
• gemini-api
• google
• google-colab
• google-drive
• google-gemini
• google-sheets
• google-workspace
• prompt-engineering
• python
• traceability