Challenges I Ran Into

Building AEGIS involved integrating AI, multi-agent systems, blockchain, and frontend visualization—each of which introduced unique challenges. Some of the most impactful hurdles are outlined below.

Blockchain Transaction Failures

The Issue:
Even after successful disaster detection and verification, Ethereum transactions were failing with errors such as underpriced replacement transactions or silent funding failures.

Root Cause:
The funding logic depended on values that were not always set correctly by upstream agents, causing invalid or zero-value transactions.

How I Solved It:
I traced the entire pipeline end-to-end, identified the missing fallback logic, and implemented defensive checks to guarantee safe transaction execution. I also resolved Web3.py version mismatches that affected raw transaction handling.

AI Confidence vs Real-World Accuracy

The Issue:
Initial computer vision detections produced inconsistent confidence scores, making it difficult to determine when an event should trigger funding.

Root Cause:
Different disaster types required different detection strategies and thresholds, which were not initially calibrated.

How I Solved It:
I separated detection pipelines by disaster category (fire, flood, structural damage, casualties) and tuned confidence thresholds individually to balance sensitivity and false positives.

Agent Coordination & State Flow

The Issue:
Synchronizing multiple agents caused race conditions where one agent acted before another completed its task.

Root Cause:
A lack of clear execution boundaries between detection, verification, and execution stages.

How I Solved It:
I introduced a message-queue-based workflow and enforced strict state transitions, ensuring that each agent only acted once the previous stage was fully completed.

Image Processing Pipeline Errors

The Issue:
Image format mismatches (bytes, PIL, OpenCV arrays) caused crashes and inconsistent results during analysis.

Root Cause:
Different libraries required different formats, and error handling was missing during conversions.

How I Solved It:
I standardized the image processing pipeline, added validation checks at each stage, and implemented graceful fallbacks to prevent system-wide failures.

Rapid Frontend Prototyping vs System Complexity

The Issue:
Displaying real-time AI decisions and blockchain transactions without overwhelming users.

Root Cause:
Complex backend processes needed to be simplified for human understanding.

How I Solved It:
I focused on clarity-first UI design—showing only essential states, using real-time status indicators, and progressively revealing technical details when needed.