Engineered a multi-agent AI system with Google's ADK and Gemini for clinical trial patient matching. Enhanced a RAG pipeline using Vertex AI Search and BigQuery for data analyses.
May 2025 - Aug 2025
Sequential Multi-Agent AI System: Engineered a stateful, sequential multi-agent framework using Google’s ADK and Gemini models to automate patient matching for oncology clinical trials. Designed an orchestration agent to coordinate specialized sub-agents for patient matching, profile generation, and eligibility filtering.
Retrieval-Augmented Generation (RAG) Pipeline & Data Infrastructure: Enhanced an end-to-end RAG pipeline integrating Vertex AI Search to retrieve and synthesize large-scale clinical data. Leveraged BigQuery for downstream analysis of agent-generated outputs, engineering robust callback mechanisms and structured JSON parsing to transform LLM responses into Pandas DataFrames for auditable analyses.
Collaborative Leadership: Authored complex prompt strategies to enforce structured, schema-consistent outputs from Gemini models, ensuring high data integrity throughout the pipeline. Co-organized 5+ intern events and led 2 initiatives fostering community engagement.
R&D Software Engineering Intern @ Chamberlain Group
Developed a smart home IoT product with low-latency video streaming solution with FFmpeg and GStreamer. Developed Python server-side code for RTSP/TCP streams and a Java/Android notification system.
Sept. 2023 - Dec. 2023
Real-Time Streaming: Designed a low-latency, scalable real-time streaming solution using FFmpeg and GStreamer, consistently maintaining high-quality 1080p playback under variable conditions, and minimal buffering to provide seamless video monitoring for security purposes.
Server-Side Development: Designed and developed server-side code in Python for RTSP and TCP streams. Achieved latency less than 1000ms from the Raspberry Pi 4 camera using the socket.io and PiCamera2 libraries. Integrated mediamtx for efficient media data management to optimize performance.
Notification System: Used Java & Android and MQTT communication Services, created 2 API endpoints to trigger alerts when the doorbell rings, achieved a success rate 100% in delivering real-time alerts.
AI & Embedded: Experimented with STMicro Sensors & YOLOv5 neural network to analyze surroundings.
UI/UX: Collaborated with a multidisciplinary team of 6 engineering, UI/UX, and product design interns to develop UI components, and conducted UX surveys and interviewed with 3 smart IoT product users.