Selected work and project archive

Quant Research Developer on the Global Stock Selection team.

Software engineering internship at Velocifile building RegDoc AI, a multi-tenant AI document processing platform for regulatory documents across 51 US states.

• Architected multi-tenant AI document processing pipeline handling 50,000+ regulatory documents across 51 states with automated ETL pipelines, real-time OCR, and hybrid semantic search achieving 1,000+ docs/hour and 28% search accuracy improvement (F1: 0.90).
• Built GPU-aware processing scheduler with adaptive concurrency control and real-time memory monitoring delivering 3.5x faster vector queries (71ms vs 247ms) and 6.1x P95 latency improvement.
• Decoupled monolith into 3 microservices for processing, search, and analytics, improving scalability by 300% and cutting deployment time by 65%.

AI-native desktop IDE built in Rust/Tauri 2 with multi-provider agentic AI, a typed IPC bridge, and human-in-the-loop approval workflows — shipping at ~15MB vs Electron's 300MB+.

• Architected 7-crate Rust workspace with 143 typed IPC commands spanning file system, terminal (PTY), git (31 commands via libgit2), code parsing, and AI agent orchestration.
• Integrated Claude Agent SDK, OpenAI, and Gemini via Node.js sidecar bridged to Rust over newline-delimited JSON on stdin/stdout, enabling persistent streaming sessions with interrupt, resume, and fork.
• Designed 3-layer AI permission system with PreToolUse hooks, canUseTool callbacks, and plan/accept mode enforcement — enabling safe agentic tool execution with human-in-the-loop approval.
• Auto-generated TypeScript bindings from 52+ Rust structs via ts-rs, ensuring compile-time type safety across the full IPC surface between Rust backend and React 19 frontend.
• Built extensible skills system and MCP server integration for dynamic tool registration, plus Tree-sitter code intelligence across 8 languages for symbol extraction and AI context building.

Built cloud-native developer tools and ML prediction servers for academic research and teaching.

• Architected cloud-native Python IDE supporting 60+ concurrent students with 99.9% uptime using AWS ECS Fargate auto-scaling and RDS PostgreSQL.
• Engineered security-hardened execution engine with thread isolation and 128MB memory limits preventing 100% of privilege escalation attacks.
• Designed real-time WebSocket architecture handling 18,000 messages/min with sub-200ms latency.
• Built FastAPI ML prediction server integrating TensorFlow for RNA splicing prediction with 10ms API response times.

Software engineering internship at Validia building Interview Guard, a remote cheating detection platform with computer vision and gaze tracking.

• Achieved 95%+ cheating detection accuracy by integrating PLGaze AI with on-screen coordinate transformation algorithms.
• Built web-based calibration system in React/FastAPI and 6-table MySQL schema, processing 1,000+ sessions with zero data loss.
• Cut onboarding time by 60% through automated installation scripts and ETL pipelines.