Harshita Raj

An agentic automation framework that converts plain-language requirement documents into fully executable low-code workflows end-to-end.

I designed the system to use Claude API for reasoning and MCP servers for tool invocation, enabling the agent to plan multi-step flows, handle decision branches, and drive a CLI to generate outputs in the target project. The pipeline goes from natural language input to 4-5 complete activity flows with branching logic, with no manual steps required. This project demonstrates real-world agentic AI engineering: prompt design, tool-calling orchestration, structured output parsing, and failure handling. It is built entirely in Python with MCP server integration.

A production internal tool that automates engineering ticket triage via Slack.

I built an event-driven back end in Node.js and TypeScript that listens for ticket events on the Slack Events API, retrieves relevant thread context using a custom MCP server, generates structured AI summaries using the Claude API, and posts threaded updates to the relevant engineering or support channels. The system replaced a manual daily triage process that required engineers to review open tickets and route them manually. The architecture includes a clean separation between ingestion, context retrieval, LLM processing, and delivery layers. All components are containerized and deployed with CI/CD.

A production-style OAuth 2.0 authorization server built in Java and Spring Boot, designed to demonstrate enterprise-grade back-end system design.

I implemented the full OAuth 2.0 authorization code flow with JWT issuance and validation, role-based access control (RBAC), Redis token bucket rate limiting, and Redis-backed token revocation. The service is fully containerized with Docker, and the GitHub Actions CI pipeline handles build, test, and packaging on every commit. All business logic is covered by JUnit tests. This project reflects the security and reliability patterns required in financial-grade and multitenant SaaS back ends—clean API contracts, stateless token validation, and layered access control.

An automated pipeline that generates draft release notes from tagged work items and routes them through a human review step before publication.

I built the system in Python using the Claude API with structured output prompting—the pipeline ingests tagged tickets via webhooks, constructs a context-rich prompt from the work item metadata, and generates formatted release notes as a structured JSON response. The output is posted to a dedicated Slack channel for engineering review and approval before going live. This eliminated a manual writing step that previously took engineers 30-45 minutes per release cycle. The pipeline is designed to be repeatable, auditable, and easy to extend with new ticket sources or output formats.