Smart Logistics Auditor

AI-powered dashboard photo audits that instantly detect mileage fraud, warning lights, fuel issues, and fleet safety risk.

Comment

Inspiration: The project was inspired by the real-world integrity gap in logistics fleets: drivers report dashboard status manually, but a single instrument cluster photo can reveal odometer truth, fuel level, and active warnings instantly. I wanted to turn that broken audit process into a reliable, AI-verified workflow.

Smart Logistics Auditor lets a fleet manager or driver upload a dashboard photo. The system uses Gemini 1.5 Flash to analyse the image and extract structured audit data:

mileage fuel percentage active warning lights safety score It stores results in SQLite and shows them in a Streamlit dashboard with audit history, fleet analytics, and vehicle management.

How we built it We built the app with:

gemini_engine.py for Gemini multimodal prompt logic and JSON extraction api.py for FastAPI endpoints, vehicle/audit management, and validation app.py for Streamlit UI covering upload, history, analytics, and fleet management database.py for SQLite persistence and CRUD helpers The Gemini call sends image bytes plus a detailed system instruction that asks the model to behave like a Senior Automotive Inspector and return only JSON.

Challenges we ran into are: Getting Gemini to output stable, parseable JSON reliably

Teaching the model precise warning-light icon mappings Handling unreadable or non-dashboard images gracefully Ensuring audit values are validated before saving, especially mileage monotonicity

Accomplishments that we're proud of: Built a full multimodal AI audit pipeline from image upload to analytics

Created a robust prompt and validation layer that enforces structured output Delivered a working Streamlit demo plus backend API and persistent SQLite storage Generated placeholder screenshot assets and complete hackathon submission docs ,

What we learned: Strong prompt engineering is essential for reliable multimodal results

Low temperature settings like (T = 0.1) improve determinism for structured JSON extraction Simple backend validation is critical when using AI-generated data A compact Python stack can support a full prototype quickly

What's next for Smart Logistics Auditor: Add real dashboard image preprocessing for better OCR and icon detection

Support additional vehicle diagnostics like tire pressure and battery health Add user authentication and fleet roles Expand analytics with trend forecasting and anomaly alerts Integrate direct mobile upload workflow for in-field inspections

Built With

  • fastapi
  • gemini1.5
  • google-genai-sdk
  • pandas-(data-analysis)-**platforms:**-local-development-environment-(windows/linux/mac)-**other:**-git-(version-control)
  • pillow-(image-processing)
  • python
  • sqlite
  • streamlit
  • streamlit-(frontend-dashboard)-**databases:**-sqlite3-(persistent-storage-for-vehicles-and-audit-logs)-**apis:**-google-generative-ai-(gemini-1.5-flash-multimodal-model)-**libraries:**-google-genai-sdk-(gemini-integration)
Share this project:

Updates