Raghavsai Tirupati
Dallas, TXragtirup@gmail.comlinkedin.com/in/raghavsaitgithub.com/Raghavsai-Tirupati
Education
- Honors: Best Use of Gemini API at TidalHack, 1st Place at Hook 'Em Hacks
Experience
Brinks Home
May 2026 – Present- Implemented C#/.NET checkout workflow changes for a platform serving 1M+ customers, including an eContract recovery that restores 200+ expired order amendments monthly and recurring-billing updates for service-plan and address changes.
- Resolved a RabbitMQ retry edge case by adding explicit completed-state checks to prevent duplicate workflow processing.
- Wrote unit/regression tests at 90% code coverage for CI/CD pipelines with sharded retries, gating production releases.
- Diagnosed workflow failures across APIs, RabbitMQ queues, and .NET services using Azure logs to isolate root causes.
ClinicalHours
Oct 2025 – Present- Launched a web platform to help pre-med students find clinical opportunities, while also helping hospitals automate volunteer onboarding/credentialing, growing to 1,000+ registered student users and 9,500+ indexed clinical facilities.
- Built a React, TypeScript, and Postgres platform with Cloudflare routing to manage clinical job postings & applications.
- Reduced clinic volunteer onboarding time by 70%+, replacing manual credentialing checklists with automated workflows.
- Built the ingestion pipeline behind the facility index, merging 88K+ federal CMS and HRSA records into 9,500+ unique facilities in Postgres via name/city/state deduplication and checkpointed, resumable enrichment runs across all 50 states.
Projects
ACSI (AI Certification System Infrastructure) | Python, LLM Evaluation, HDBSCAN
- Built an open-source tool that replays real LLM traffic to evaluate model swaps and return pass/block verdicts.
- Caught a swapped model silently breaking the JSON contract on 53% of 279 real traces, versus 0 for the original model, while blind judges (separate LLMs) still scored the new model's answers as good or better on every pair.
- Measured a 3.9% noise floor, clustered 149 failures with HDBSCAN, and validated a prompt fix by re-certifying (149 → 18 → 5 failures), with the certifier correctly withholding auto-pass on the residual five.
Iris, 1st Place at Hook 'Em Hacks | Swift, React, TypeScript
- Shipped a first-place iOS computer-vision application that turns live camera input into speech for blind users.
- Integrated a multimodal LLM agent to answer follow-ups about users' surroundings with context-aware speech.
- Added an OCR reading mode for signs, menus, books, and labels, returning spoken text output in under 4 seconds.
Matchbook Trade Engine | Java, Spring Boot, PostgreSQL, JUnit, JMH, Docker
- Created a Java and Spring Boot trading engine that matches buy and sell orders through backend APIs.
- Designed client-server architecture with REST/WebSocket APIs, PostgreSQL storage, Docker, and CI.
- Tested order-matching logic with JUnit and JMH to validate correctness, latency, and edge-case behavior.
Technical skills
Languages: Python, TypeScript, Go, Java, C++, SQL, C#, Swift, R
Frameworks & Libraries: Spring Boot, .NET, Node.js, React, Next.js
Tools: Linux, Docker, AWS, Azure, Git, MySQL, PostgreSQL, REST APIs
Development: Agile/Scrum, unit & regression testing, CI/CD, REST API design, Jira