Important
Hackathon submission note: The first ~38 seconds of the demo video appear blank due to a recording/export issue. The actual demo begins at 00:38.
Technical reasoning, live logs, and the dashboard deep-dive begin at 00:38
Trace-Ability acts as a Cognitive Black Box Recorder for software architecture. Instead of only storing code diffs, it captures the reasoning and architectural intent behind each developer commit.
The system analyzes modified files, reconstructs the likely purpose of the change, and stores these insights as structured telemetry - creating a traceable record of how the system evolves.
| The Flaw | The Catalyst | The Reality |
|---|---|---|
| Git tracks what changed, but erases the why. | AI accelerates code generation without capturing intent. | Endless hours wasted reverse-engineering massive technical debt. |
Six months later, developers cannot explain why a critical service dependency was introduced . Trace-Ability bridges this gap by generating living, evolving documentation natively through semantic analysis, avoiding reliance on misleading or incomplete human commit messages .
Trace-Ability uses Amazon Nova Lite to analyze raw code diffs and reconstruct the architectural intent behind each commit .
- Automated Intent Reconstruction: Performs semantic analysis on code diffs to generate living documentation of architectural decisions.
- Immutable Telemetry Ledger: Stores project intelligence in a cryptographically verifiable log backed by Amazon DynamoDB.
- Cognitive Value Add: Generates trust scores and risk signals, turning complex diffs into concise summaries to reduce code-review effort.
To reduce the “black-box” uncertainty common in AI tooling, Trace-Ability assigns a deterministic trust score to each generated output using three core signals.
- Model Confidence: The LLM confidence score for the generated architectural explanation.
- Diff Consistency: Measures how well the commit message aligns with the actual code changes.
- Change Locality: Checks whether modifications stay within a logical module or are scattered across the codebase.
Trace-Ability operates on an event-driven CI pipeline built on GitHub Actions and powered by AWS Serverless technologies .
Figure 1.0: End-to-End Serverless Intelligence Layer
- GitHub Repository: Entry point where commit and push events trigger the webhook.
- AWS Lambda (Webhook Receiver & Processor): Processes GitHub events and extracts raw diffs and metadata.
- Amazon Bedrock (Nova Lite): Performs semantic analysis to generate the architectural narrative.
- Amazon DynamoDB (Telemetry Ledger): Stores commit intelligence and analysis results as an immutable record.
- Next.js Dashboard: Frontend deployed on Vercel that retrieves data via Lambda Function URLs to display trust scores and real-time updates.
Figure 1.1: Multi-Stage Telemetry Ingestion and Agentic Reasoning Flow
- Developer Action: A developer commits and pushes code to the repository.
- Event Capture: Raw diff and commit metadata are intercepted and extracted.
- AI Analysis: The system determines intent and calculates risk and trust scores.
- Ledger Storage: The structured architectural narrative and scores are stored securely.
- Dashboard Visualization: Results are streamed to the dashboard for real-time review.
The intent-extraction pipeline runs on a fully serverless architecture, enabling efficient scale-to-zero operation.
- Ultra-Low Deployment Cost: The entire system - development, testing, and deployment - ran on approximately $0.10 total cloud cost, demonstrating the efficiency of a serverless AI architecture.
- Cost Optimization: The intelligence engine runs on Amazon Nova Lite, keeping the estimated MVP operating cost below $10/month.
- Semantic Accuracy: Evaluated on a ground-truth dataset of 50 complex commits, achieving an 81.4% semantic match rate between the generated narrative and human intent.
- Inference Latency: Average intent-extraction time < 3.5 seconds, enabling CI/CD integration without blocking workflows.
- Scale & Complexity: Processed average diffs of 8–25 files per commit, with the largest test handling 42 files.
| Category | Stack |
|---|---|
| Cloud Compute & API | AWS Lambda, Lambda Function URLs |
| Generative AI | Amazon Bedrock (Amazon Nova Lite) |
| Database | Amazon DynamoDB (with secured IAM least-privilege Scan/Query policies) |
| Frontend Framework | Next.js 15 (App Router), React, Tailwind CSS |
| Animations & UI | Framer Motion (Shared Layout Transitions, 3D Horizon Scrolling), Lucide Icons |
| DevOps & Deployment | Vercel (Custom Root Directory configs), GitHub Actions |
FIGURE 1:Operational Command Center: Real-time Cognitive Engineering HUD
FIGURE 2: Neural Narrative: Semantic Intent Reconstruction & Confidence Scoring
FIGURE 3: System Initialization: Bedrock-Enforced Telemetry Scanning
FIGURE 4: Trace-Ability Core: Cognitive Black Box Recorder Branding
Repository structure overview.
.
├── README.md
├── PRODUCT_SPECS.md
├── design.md
├── requirements.txt
│── architecture.png
│── flowdiagram.png
├── assets
│ ├── icons
│ │ ├── DynamoDB.svg
│ │ ├── Lambda.svg
│ │ ├── ai.svg
│ │ ├── api.svg
│ │ ├── brain.svg
│ │ └── ...
│ └── screenshots
│ ├── screenshot_1.png
│ ├── screenshot_2.png
│ ├── screenshot_3.png
│ ├── screenshot_4.png
│ ├── screenshot_5.png
│ └── screenshot_6.png
├── backend
│ └── lambda_handler.py
├── dashboard
│ ├── package.json
│ ├── next.config.ts
│ ├── tailwind.config.ts
│ └── src
│ ├── app
│ └── components
├── hooks
│ └── pre-commit.py
├── data
│ └── schema.json
└── tests
├── github_api_test.py
└── sample_diff.json
Trace-Ability runs on AWS Lambda, so the pipeline executes only when a git push triggers it.
Bash
# 1. Clone the repository
git clone [https://github.com/sohamrajput98/Trace-Ability.git](https://github.com/sohamrajput98/Trace-Ability.git)
cd Trace-Ability
# 2. Configure Local Webhooks
chmod +x hooks/pre_commit.py
ln -sf ../../hooks/pre_commit.py .git/hooks/pre-commit
# 3. Environment Provisioning
pip install -r requirements.txt
# Ensure AWS CLI is configured with Bedrock (Nova Lite) permissions
# Deploy backend/lambda_handler.py to AWS Lambda
-
Multi-Signal Context Aggregation: Expanding the pipeline to ingest Issue descriptions (Jira) , PR discussions, and Slack threads alongside the code diff to generate a complete 360-degree view of developer intent .
-
Pull Request Advisor: Injecting the Bedrock architectural narrative directly into GitHub PR comments as an automated reviewer before code merge .
-
Semantic "Why" Search: Allowing developers to query the DynamoDB intent-registry (e.g., "Why did we introduce Redis to the auth service in January?") .
