Redacta
ArchivedRedacta tackles the tedious, error-prone task of manually redacting personally identifiable information from documents. Users upload files through a React/TypeScript front-end; a Spring Boot back-end orchestrates a LangChain-powered LLM pipeline that detects and anonymises names, addresses, and other PII in seconds. The platform is packaged with Helm charts and deployed on a Rancher-managed Kubernetes cluster with an AWS-based CI/CD pipeline (CodePipeline + ECR). I built the back-end API, integrated the LangChain pipeline, wrote Helm charts for multi-environment deployment, and contributed to the React UI.
Key Impact
- Automated PII detection and anonymization
- LLM-powered intelligent document processing
- Cloud-native deployment on Rancher
Features
- →LangChain-based PII detection pipeline with configurable entity types
- →Batch upload and processing of PDF, DOCX, and plain-text files
- →Audit trail and versioned redaction history
Challenges
- →Fine-tuning LLM prompts for high-recall PII detection across document formats
- →Designing a Helm chart that works across dev, staging, and production Kubernetes namespaces
- →Managing large file uploads and streaming responses between front-end and back-end