Software Engineer & Founder
Agentic Systems • Distributed Backends • AI Developer Tooling
M.S. Computer Science (AI/ML), Duke University

I am a Software Engineer and M.S. Computer Science (AI/ML) student at Duke University, focused on building production-grade, AI-powered systems with end-to-end ownership. My work sits at the intersection of agentic software architectures, distributed backends, and developer-facing tooling, where correctness, performance, and reliability matter as much as model capability.

I am the sole founder and engineer of VYNN AI, an agentic financial analyst platform built entirely end-to-end, spanning backend services, multi-agent orchestration, real-time market data pipelines, and LLM-driven analysis to deliver actionable investment insights in production. Previously, I designed and built AutoCodeRover, an LLM-based autonomous code repair system acquired by SonarSource, integrating agentic reasoning directly into the IDE workflow. Across research and industry, I focus on turning AI ideas into scalable, observable, and maintainable software systems.

🔗 https://vynnai.com
Founder & Sole Engineer

• Built and deployed a full-stack, production agentic system for financial analysis and valuation with end-to-end ownership
• Designed multi-agent orchestration using LangGraph and tool-augmented reasoning, enabling autonomous analysis, Self-RAG, and structured report generation
• Implemented distributed, async backends (FastAPI, Redis, MongoDB) with real-time SSE/WebSocket streaming
• Deployed containerized services on cloud infrastructure using Docker and Caddy, supporting CI/CD, persistent storage, and fault tolerance
• Optimized parallel execution and result caching, reducing end-to-end latency by 72% in production

🔗 https://www.sonarsource.com/company/press-releases/sonar-acquires-autocoderover-to-supercharge-developers-with-ai-agents/
Software Engineer

• Improved autonomous code repair performance to 46% SWE-bench Verified and 37% Lite through algorithmic and systems-level enhancements
• Built a JetBrains IDE plugin in Kotlin, integrating with Sonar tooling for real-time diagnostics and autonomous repair
• Designed AST-level Patch Alignment using GumTree to reconcile LLM-generated patches with live developer edits
• Implemented a Self-Repair Agent with feedback-driven replay loops and structured state tracking
• Reduced manual debugging effort by ~40% through IDE-native agentic workflows

AI Researcher

• Built a scalable multi-agent framework for large-scale citation screening and text mining
• Evaluated across 150K+ abstracts from 15 systematic reviews
• Achieved 99.5% sensitivity and 87.9% specificity, with 99.95% NPV, relative to manual screening baselines
• Primary contributor to manuscript under review at NEJM AI (2025)

• Graduate Teaching Assistant, Duke University — Software Delivery & Software Development
• Founder & Software Engineer, VYNN AI
• Software Engineer, AutoCodeRover (acquired by SonarSource)
• Software Engineer, Binance — large-scale API testing and reliability
• Quantum Software Engineer, Center for Quantum Technologies (NUS)
• Full-Stack Software Engineer, ST Engineering — real-time systems and CI/CD

Languages
Python, Kotlin, Java, TypeScript/JavaScript, C#, SQL, R

Systems & Backend
Distributed services, async architectures, FastAPI, REST APIs, Docker, Redis, MongoDB, CI/CD, Linux, multithreading & concurrency

AI Systems & Tooling
Agentic workflows, LLM tooling, SWE-bench, Self-RAG, text mining, evaluation pipelines, developer-facing AI systems

Testing & Reliability
API testing (Postman), web automation (Selenium), load & stress testing (JMeter), end-to-end and regression testing

• Email: zanwen.fu@duke.edu
• LinkedIn: https://www.linkedin.com/in/zanwenfu

_{Last updated: Aug 2025}