Building Glean's agentic business observability platform

• Engineered AWS-managed Airflow to clean and transform commercial lease PDFs into LLM-generated Question-Answer (QA) pairs using GPT-4o, Pinecone and LangChain RAG to achieve 5x increase in lease processing throughput for 80% reduction in product pricing.

• Led development of Next.js and React TypeScript UI for validating QA pairs with human feedback, doubling user feedback score to 99%.

• Optimized platform operations by integrating GitHub Actions for CI/CD and unit testing, Twilio for OAuth, Stripe Connect for payouts, and Slack API for in-platform support, slashing administrative hours for the Chief Operating Officer by 60%.

• Designed, deployed, and maintained high-availability B2B platform in Next.js, React TypeScript, Jest, and Kubernetes for carbon accounting and offsetting, handling $60M in credits for 200+ enterprises from 340 globally-sourced sustainability projects.

• Centralized accounting and offsetting services into serverless, event-driven REST API provisioned in Terraform and AWS to enhance scalability and maintainability, yielding 45% reduction in operating costs, 30% faster release cycles, and 99.8% uptime.

• Identified Carbonlink API as growth driver by analyzing log data in Elasticsearch dashboard, leading to 125% surge in monthly revenue.

• Exceeded product timeline expectations with 100% sprint completion rate for 12-person team by meticulously detailing feature requirements and automating CI/CD and user story management with GitHub Actions.

Awards:

• USC TroyLabs Startup Accelerator "Startup of the Year"

• Contrary Venture Capital Group's Nationwide College Startup Championship

• USC Lloyd Greif Center for Entrepreneurial Studies Startup Accelerator Participant

• Spearheaded comparative analysis on 7 convex hull algorithms within parallel and distributed environments, benchmarking performance metrics like runtime and scalability to evaluate their applicability for large-scale data processing.

• Adapted and optimized Graham Scan algorithm for GPU acceleration with CUDA, NumPy, and Shapely for brain signal pathway reconstruction, reducing processing time from 23 minutes to 1 minute for near real-time simulation of pathway geometries.

• Rectified bit overflow in flight navigation code isolated through rigorous unit testing in C++, saving $1M+ from averted unit recalls.

• Tuned ARIMA model for denoising accelerometer signal input, isolating original signal with 99% accuracy after 100ms.