Ship the AIRun the platform
AI-Assisted Secrets Detection for Large Codebases
Developed an AI-powered secrets detection tool that processes large codebases into manageable chunks, calls an internal AI model trained on company-specific sensitive terms and credential patterns, and generates detailed reports to prevent credential leaks.
- client
- Qualitest Group
- when
- 2023-2024
- headline
- 75% fewer hard-coded secrets
01 · Problem
Generic secrets scanners (Gitleaks, TruffleHog) miss company-specific patterns and generate high false-positive rates. AI-driven detection with custom training on internal data reduces false negatives, catches context-specific secrets, and prevents credential leaks before they reach production.
02 · What I built
- Built codebase chunking system to process large repositories efficiently
- Trained internal AI model on company-specific secrets, credentials, and sensitive data patterns
- Developed API integration for AI model inference and response aggregation
- Implemented automated report generation with severity scoring and remediation guidance
- Integrated into CI/CD pipelines (Azure DevOps, GitLab CI, GitHub Actions) as security gates
- Co-developed LLM usage proxy and Chrome extension for sensitive-content checks on ChatGPT/Claude
03 · Result
- Reduced credential leaks and hard-coded secrets by 75%
- AI-trained on company-specific sensitive patterns
- Automated secrets detection in CI/CD pipelines
- Real-time LLM guardrails for developer tools