Supply Chain Intelligence System

Google Antigravity Hackathon — Challenge 1

---

Project Overview

An agentic AI system that ingests multiple logistics data sources, detects contradictions, generates a constrained action plan, simulates execution, and shows measurable outcome changes.

---

Domain

Logistics / Supply Chain

---

Architecture

The system uses 4 agents orchestrated via Google Antigravity:

1. Content Ingest Agent Reads 5 data sources and extracts signals with credibility scores

2. Insight & Contradiction Detector Identifies conflicting sources, flags stale data, generates core insight and risk level

3. Action Planner Agent Generates 5-step action chain within PKR 500,000 budget constraint

4. Simulation Agent Executes actions, handles failures, shows before vs after state change

---

How Google Antigravity Is Used

• All 4 agents run as separate conversations in Antigravity Agent Manager
• Each agent receives output of previous agent as input (orchestrated chain)
• Plan mode used to generate full reasoning traces and artifacts
• Agent Manager logs show complete decision trail for all 4 agents

---

Data Sources (5 inputs)

1. warehouse.csv — stock levels (structured)
2. supplier_email.txt — delay notice (unstructured)
3. sales_dashboard.json — demand data (semi-structured)
4. complaints.csv — customer sentiment (structured)
5. news_feed.txt — external risk (unstructured/real-time)

---

Tools and APIs Used

• Google Antigravity — agent orchestration
• Flutter — mobile app framework
• Dart — programming language

---

Constraints Applied

• Budget: PKR 500,000 maximum
• Urgency: All actions within 48 hours
• Actions: Maximum 5 in chain

---

Assumptions

• Data files are mock/simulated for demo
• No real personal data used
• WhatsApp API failure is intentionally simulated to demonstrate recovery

---

Failure and Recovery Evidence

ACT-002 WhatsApp API notification failed with error 503. System automatically triggered SMS fallback. Recovery successful. See failure_log.json for full details.

---

Baseline Comparison

A simple non-agentic system would only summarize the 5 inputs. Our system goes further by detecting contradictions between sources, applying credibility scoring, generating a constrained action chain, simulating execution with failure recovery, and showing measurable state change.

---

Cost and Scalability

• Cost per agent run: approximately $0.02
• 10x scale: system handles 50 sources with same architecture
• 100x scale: parallel agent conversations in Antigravity handle 500 sources

---

Limitations

• Data sources are mock, not live APIs
• Mobile app runs as web app in browser
• Action simulation is not connected to real external systems

---

Team

[Team Name: CodeWeaver] [Members Name: Irfan Karim, Haider Anjum]

Downloads

Full app and demo video available here: Google Drive - CodeWeaver Submission