EduShare Agent

Autonomous Inter-School Surplus Redistribution Network

Comment

EduShare Agent — Autonomous Inter-School Surplus & Logistics Redistribution Network

Inspiration

Every academic year, an invisible contradiction unfolds across public schools: in one school, a storage room sits packed with idle desktop computers, science lab microscopes, or robotics kits gathering dust. Meanwhile, a neighboring school is canceling its computer programming club or science laboratory experiments due to acute budget and equipment shortages.

Why doesn't this surplus move? Because public school administrations face intense administrative friction. School principals are already overburdened with daily operational duties, and navigating bureaucratic paperwork, finding which school needs what, coordinating transport logistics, and ensuring formal accountability is nearly impossible to do manually.

We asked a bold question: What if schools didn't have to search, negotiate, and fill out endless forms? What if an autonomous multi-agent system could proactively discover dormant assets, evaluate urgent educational needs, calculate optimal logistics routes, and orchestrate legal transfers—all while keeping school principals securely in the decision-making loop?

This question gave birth to EduShare Agent: Turkey's first autonomous, AI-driven inter-school surplus and logistics redistribution network.

What it does

EduShare Agent connects schools into an intelligent, circular resource-sharing ecosystem:

  1. Multimodal Visual Asset Ingestion (Amazon Bedrock Nova Pro & Claude 3.5 Sonnet): A school principal or inventory officer snaps a single photo of unused equipment or an entire storage room. The vision agent automatically identifies the equipment, categorizes it (IT, Science Lab, Furniture, Sports, Arts, Library), assesses physical wear and condition, estimates market replacement value, and indexes it into the school's surplus registry in under three seconds.

  2. Autonomous Multi-Agent Matchmaking Engine: The matchmaking agent continuously operates in the background, matching open school needs against available surplus inventory. It calculates multi-factor utility scores factoring in:

    • Geographic Proximity: Hyper-local distance calculations via the Haversine formula to minimize transit time and carbon footprint.
    • Urgency Weighting: Prioritizing critical classroom shortages (e.g., coding exams, laboratory requirements).
    • Condition Compatibility: Matching educational equipment standards with the target school's grade level and facility.

  3. Bilateral Handshake with Human-in-the-Loop (HITL) Governance: In the public sector, autonomous systems cannot make unilateral decisions about institutional property. EduShare Agent introduces a rigorous Bilateral Handshake Protocol:

    • The AI agent formulates an official transfer recommendation.
    • Both the donor school principal and the recipient school principal receive a prompt in their Agent Decision Center.
    • Both schools review the condition, distance, and quantity, and must explicitly approve or modify the transfer.
    • Only when mutual consensus is reached does the transfer enter the dispatch state.

  4. Live Geospatial Map & Real-Time Event Streaming: An interactive Leaflet-powered command dashboard provides a live map of regional schools, inventory pins, and animated transfer routes. Real-time Server-Sent Events (SSE) stream agent decisions, status updates, and peer requests across all connected browser sessions without requiring page refreshes.

  5. Measurable ESG & Financial Impact Analytics: Every successful reallocation dynamically updates real-time impact indicators:

    • Public Budget Saved (₺/$): Capital expenditure avoided by reusing existing public inventory.
    • CO₂ Emissions Prevented (kg CO₂e): Environmental savings calculated from avoiding new manufacturing emissions and optimizing local delivery routes.
    • Active Redistribution Ratio: Measuring the velocity of idle public assets returning to active learning environments.

  6. Enterprise-Grade AI Quota & Billing Circuit Breaker: To ensure public demo reliability and eliminate unexpected cloud expenses, an intelligent billing shield enforces a daily circuit breaker (300 requests/day), per-school quotas, sliding-window rate limits, and SHA-256 visual caching (identical image uploads return cached results in 2ms with zero token cost).

How we built it

EduShare Agent was architected from the ground up as a production-ready, cloud-native system:

  • AI & Vision Engine:

    • Amazon Bedrock (Nova Pro & Claude 3.5 Sonnet): Powers multimodal visual analysis, structured JSON extraction, and autonomous agent reasoning.
    • Custom prompt schemas enforce strict extraction of item titles, standardized categories, condition ratings, and unit values.

  • Backend Architecture:

    • FastAPI (Python 3.11 ASGI): Asynchronous, high-throughput REST API and Server-Sent Events (SSE) broadcaster.
    • SQLAlchemy & PostgreSQL 16: Relational schema modeling schools, inventory items, need requests, transfer state machines, and audit trails.
    • Bilateral State Machine: Strict status transitions (PROPOSED → AWAITING_DONOR_APPROVAL → AWAITING_RECIPIENT_APPROVAL → IN_TRANSIT → COMPLETED).
    • Security Shield: Custom rate limiters, perceptual image hashing, 4MB file safety checks, and JWT-based role authentication.

  • Frontend & User Experience:

    • React 19 & Vite: Ultra-fast, component-driven client application built for desktop and mobile devices.
    • Tailwind CSS & Framer Motion: Clean, modern interface with micro-interactions, responsive mobile drawer navigation, and accessible typography.
    • Leaflet & React-Leaflet: Geospatial mapping with custom SVG markers, dynamic clustering, and animated polyline transfer arcs.
    • i18next Localization: Complete bilingual support (Turkish & English) across all notifications, metrics, and modals.
    • PWA (Progressive Web App): Service worker caching and installable mobile experience for school staff on the move.

  • Cloud Infrastructure & DevOps:

    • AWS EC2 (Ubuntu 24.04 LTS): Production deployment hosting containerized backend, database, and web server.
    • Docker & Docker Compose: Multi-container architecture orchestrating edushare-backend, edushare-frontend, and edushare-postgres.
    • Nginx Reverse Proxy: Efficient routing, gzip compression, and static asset delivery.
    • GitHub Actions CI/CD: Automated testing, container builds, and zero-downtime SSH deployment on push to main.

Challenges we ran into

  1. Balancing Autonomous Agency with Public Accountability: Creating an autonomous agent for public schools required careful boundary definition. An AI cannot legally transfer state property on its own. Designing the Bilateral Handshake (Human-in-the-Loop) protocol required building a resilient state machine where the agent takes care of 95% of the coordination, discovery, and logistics planning, while leaving the final 5%—the authoritative sign-off—firmly in the hands of authorized school administrators.

  2. Real-Time Geospatial Map Synchronization: Rendering dozens of schools, dynamic surplus/need badges, and animated transfer routes while maintaining 60 FPS on mobile browsers was challenging. We solved this by decoupling map marker state from high-frequency telemetry, utilizing lightweight SSE streams for push notifications, and implementing efficient SVG canvas rendering for transfer route polylines.

  3. Cloud Cost Protection on Public Demo Endpoints: Publishing an AI-powered multimodal platform for a hackathon jury means anyone can test the system. Unrestricted image analysis calls to advanced LLMs could easily cause runaway API bills. We engineered an intelligent multi-layer Billing Shield:

    • SHA-256 cryptographic image hashing returns instant cached results for duplicate uploads at 0 token cost.
    • Automatic client-side image compression (max 800x800 JPEG).
    • Global circuit breaker capping daily Bedrock calls at 300, with dedicated school quotas and guest trial limits.

  4. Securing Educational Operations from Unauthorized Input: Early in testing, we noticed guest visitors could open surplus and need forms, which led to confusion when unauthenticated submissions were rejected by the backend. We designed an intuitive dual-layer auth gate: proactive login prompts on action triggers, combined with in-modal visual auth shields that explain why MEB school verification is required for public inventory transactions.

Accomplishments that we're proud of

  • ⚡ Complete End-to-End Autonomous Loop: From taking a smartphone photo of idle equipment in a storage room to autonomous Bedrock identification, agent matchmaking with a school in need, bilateral approval in the HITL center, and live route visualization on the map—all working seamlessly in production.
  • 🛡️ Zero-Waste AI Architecture: Achieving sub-millisecond cached responses for repeated vision scans, proving that AI-driven applications can be both powerful and financially sustainable.
  • 🌍 Tangible ESG & Economic Impact: Building real, quantifiable tracking for taxpayer money saved and CO₂ emissions avoided, proving the tangible value of a circular education economy.
  • 📱 Sleek, Responsive UI Across All Screen Sizes: A carefully engineered desktop navbar that collapses into a lightweight mobile drawer and bottom navigation bar, making the platform accessible on any device from classroom tablets to 27-inch desktop monitors.
  • 🚀 100% Live AWS Deployment: Not just a localhost prototype. EduShare Agent is fully deployed on AWS EC2, containerized with Docker, wired to GitHub Actions CI/CD, and live for the jury to evaluate.

What we learned

  • Agentic AI thrives on structure: Large Language Models are most effective when paired with deterministic state machines and clear domain constraints. Giving an agent unbounded freedom creates unpredictability, but giving it clear parameters (distance, urgency, category rules, bilateral sign-offs) allows it to excel.
  • Human-in-the-Loop (HITL) builds institutional trust: In public administration, users don't want AI to replace human judgment—they want AI to eliminate tedious legwork and present clear, actionable decisions.
  • Multimodal LLMs are transformative for inventory management: Traditional asset management requires tedious manual data entry (typing specs, serial numbers, categories). Amazon Bedrock's visual comprehension transforms a 5-minute data-entry chore into a 3-second snapshot.

What's next for EduShare Agent

  1. Autonomous Courier & Logistics Integration: Partnering with municipal courier services and smart delivery lockers to automatically dispatch pickup requests as soon as bilateral approval is granted.

  2. Predictive Curriculum-Based Demand Forecasting: Analyzing the national academic calendar (e.g., upcoming STEM project weeks, exam periods, robotics competitions) to predict which schools will need specific equipment weeks before the teachers even submit a request.

  3. Digital QR & NFC Asset Tagging: Automatically generating printable, MEB-compliant QR codes upon transfer approval, allowing recipient schools to scan and verify equipment intake instantly upon delivery.

  4. Multi-District & Nationwide Expansion: Expanding the agent clustering network from local pilot districts to regional hubs across all 81 provinces, creating a nationwide circular economy for educational infrastructure.

Built With

Share this project:

Updates

Submission history