Inspiration
Modern engineering teams rely on multiple sources of truth — PRDs, meeting notes, and source code.
However, verifying whether requirements are actually implemented requires hours of manual review by senior engineers.
I wanted to explore whether AI could automatically reason across these documents and detect requirement mismatches — not through keyword search, but through deep semantic understanding.
This led to building the Cross-Document Insight Engine, powered by Gemini 3.
What it does
Cross-Document Insight Engine is an AI-powered analysis tool that detects inconsistencies between product requirements and real implementation.
Supported Inputs (Version 1)
• Product Requirements Document (PDF)
Contains product goals, functional requirements, assumptions, and constraints.
• Source Code Repository (GitHub URL)
Represents the real implementation of the product.
• Meeting Notes (PDF or TXT)
Captures design discussions, trade-offs, and decisions.
Core Question It Answers
“What assumptions in the PRD are violated by the current codebase?”
Output Generated
• Assumption
• Classification — Matched / Missing / Incorrect
• Evidence from PRD, Code, and Meeting Notes
• Risk
• Suggested Fix
How we built it
The system works in four stages:
- Document ingestion and text extraction from PRDs and meeting notes
- Repository cloning and source code scanning
- Context aggregation across all inputs
- Gemini 3 analysis for structured engineering reasoning
Gemini 3 is critical for:
• Long-context reasoning across thousands of lines of code
• Cross-document synthesis
• Structured JSON engineering outputs
Challenges we ran into
• Handling large context across multiple document types
• Designing prompts that produce structured engineering analysis
• Ensuring output is explainable and evidence-backed
• Balancing scope — focusing on one high-value engineering question instead of building a generic chatbot
Accomplishments that we're proud of
• Built a real multi-document AI engineering analysis pipeline
• Generated structured outputs useful for real engineering workflows
• Demonstrated real-world use of long-context AI reasoning
• Created a tool that can potentially reduce hours of manual engineering review
What we learned
• Long-context AI can solve real engineering workflow problems
• Structured output design is as important as model intelligence
• Prompt engineering matters when building production-style AI tools
• Real value comes from solving specific high-impact problems
What's next for Cross-Document Insight Engine
• CI/CD pipeline integration
• Pull request review automation
• Support for private repositories
• Team dashboards for requirement tracking
• Real-time engineering compliance monitoring
Mission:
Replace hours of manual senior engineering review with explainable AI-driven insight.
Built With
- ai-driven-multi-document-reasoning
- and-cross-artifact-semantic-analysis-document-&-code-processing-pipeline:-multer-(multipart-file-ingestion)
- backend-orchestration-pipeline
- css
- css3-ai-&-intelligence-layer:-gemini-3-api-for-long-context-multi-document-reasoning
- custom-github-repository-source-code-scanner-and-aggregator-infrastructure-&-dev-tooling:-dotenv-for-secure-environment-configuration
- custom-meeting-notes-text-parser
- custom-repo-parsing-pipeline
- dotenv
- express.js
- express.js-rest-api-architecture
- gemini-3-api
- git-+-github-for-version-control-and-collaboration-system-architecture-concepts:-ai-augmented-static-analysis
- github-api
- html
- html5
- javascript
- multer
- multi-source-context-synthesis
- node.js
- node.js-runtime
- npm
- npm-ecosystem-for-dependency-management
- pdf-parsing-libraries
- pdf-parsing-libraries-for-prd-extraction
- pdf-text-extraction
- rest-apis
- rest-architecture
- structured-json-response-generation
- vanilla-javascript-frontend
