Commit 1caa57

2025-12-02 07:33:08 Shilpa: Updated document
Projects/Audit-matic.md ..
@@ 7,7 7,7 @@
**Team:** [[https://wiki.enovate-it.com/Teams/Team%20Fusion]]
- **Project Status:** Planning
+ **Project Status:** Development/Scaling (Actively implementing the RAG pipeline for Content Validation)
---
@@ 22,13 22,20 @@
> ### **The Problem**
- CMMI compliance auditing requires auditors to read through complex business documents and manually verify alignment with tickets in Jira and commits in GitLab. This process is slow, costly, and prone to human error, which can result in failed audits and non-compliance.
+ CMMI compliance auditing requires auditors to **manually verify alignment and content** across complex business documents and development tools. This process is:
+ * Slow, costly, and resource-intensive (Phase 1 focus).
+ * Prone to human error and subjective interpretation (Phase 2 focus).
+ * Scales poorly across large organizations with diverse toolsets (Phase 3 focus).
> ### **Our AI Solution**
**What does it do?**
- AuditMatic is a web application that automatically checks CMMI compliance by cross-validating uploaded business requirement documents against live data from Jira and GitLab.
+ AuditMatic transforms the audit process into a fast, reliable, and AI-driven pipeline by integrating:
+ 1. **Phase 1:** Automated document existence and basic entity checks.
+ 2. **Phase 2:** Deep **Content Validation** using the **RAG Pipeline** and the Gemini LLM.
+ 3. **Phase 3:** **Cross-Platform Ecosystem Coverage** (Jira, GitLab, Bitbucket) for universal traceability.
+
**How does it work?**
@@ 47,21 54,22 @@
> ### **Technology Stack**
- **AI/ML Models:** spaCy (NLP/NER), Custom Rule-Based Expert System
+ **AI/ML Models:** Gemini API (gemini-2.5-flash, embedding-001), OpenAI API (gpt-4o-mini, text-embedding-3-large), Xenova/Transformers (Local Embeddings), Custom Rule-Based Expert System, spaCy (Pre-processing)
+
+ **LLM:** Gemini (Google) and GPT (OpenAI), managed via a multi-LLM Strategy Pattern.
**Frameworks & Libraries: ** jira, multer
- **Backend:** Express.js (Node.js)
+ **Backend:** Node.js v24.x with Express.js v5.1.0 (Built with TypeScript), ts-node, nodemon.
- **Frontend:** React
+ **Frontend:** React v18.2.0 (Built with TypeScript), Redux Toolkit (RTK) (State Management).
- **Databases:** N/A (Pilot Phase)
+ **Databases:** PostgreSQL 14+ (Primary DB), pgvector v0.2.1 (Vector Search Extension), Redis v5.9.0 (Caching, Session Mgmt).
- **Deployment & Tools:** Git, GitHub, Docker
+ **Deployment & Tools:** Git, GitHub, Docker, Jest (Testing), ESLint (Linting).
- **APIs Used:** Jira API, GitLab API
+ **APIs Used:** Jira API (Phase 3), GitLab API (Phase 3), Bitbucket API (Phase 3), Google Drive API, Google OAuth 2.0 API.
- **LLM: **Gemini OpenAI
> ### **Project Architecture**
@@ 85,11 93,22 @@
> ### **Future Roadmap**
- **Short-term Goal: **Implement core compliance checks (Traceability, Accountability, Completeness) using regex-based entity recognition.
+ **Phase 1: Foundation & Entity Extraction (Completed)
+ Goal: Prove the concept of automation by replacing manual document checks.
+
+ Key Achievement: Implemented basic checks for document existence using simple Regex and spaCy for entity extraction.
+
+ **Phase 2: Content Validation & AI Pivot (Current Focus)
+ Goal: Move from "existence" to "content validation."
+
+ Key Achievement: Developed and implemented the RAG pipeline and Multi-LLM Strategy to semantically validate document content against live data.
+
+ **Phase 3: Ecosystem Expansion & Automation (Future Roadmap)
+ Short-term Goal (P3): Integrate the Bitbucket API for unified code traceability.
- **Medium-term Goal:** Train a custom NLP model to recognize non-standard internal project identifiers for better accuracy.
+ Medium-term Goal (P3): Implement AI-driven Gap Filling (LLM suggests missing content or remediation steps).
- **Long-term Goal:** Use machine learning to automatically categorize different types of requirement documents (e.g., Test Plan, Risk Plan, Functional Spec) and apply different rule sets.
+ Long-term Goal (P3): ML-based Document Categorization for dynamic rule application across different document types.
> ### **Repository & Live Demo**
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9