AI Coding Assistant @ Roche
Designed and shipped a SWARM-based multi-agent system with RAG-powered documentation retrieval for clinical trial data scientists. Achieved 1,300+ queries from 260+ users within 3 weeks of launch.
Summer Deng
AI Product Manager, builder, and cat enthusiast based in Toronto.
scroll to explore
About Me
Hi! I'm Summer, an AI Product Manager with 3+ years of experience leading enterprise AI products from discovery to launch. I've built LLM-powered multi-agent systems at Roche and owned computer vision products at Tencent generating 20M+ CNY revenue.
My approach is collaborative and iterative. I believe the best AI products emerge when cross-functional teams — ML researchers, engineers, and product — are aligned around shared purpose and empowered to experiment. Whether I'm designing evaluation frameworks, running customer discovery, or driving adoption through strategic integration, I focus on removing friction and amplifying impact.
When I'm not thinking about product, you'll likely find me with my cats Cookie and Pumpkin, exploring trails around Toronto, or diving into the latest developments in AI. I'm always curious about new ideas and love connecting with fellow builders and thinkers.
AI Product Management
LLM / Multi-Agent Systems
Cross-functional Leadership
ML Evaluation Frameworks
User Research
Computer Vision
Selected Works
Enterprise Face Recognition @ Tencent
Led product development for an enterprise SaaS face recognition system serving 100K+ daily users. Owned full product lifecycle for on-premises deployment, generating 20M CNY revenue in the first year.
Cloud Migration Automation @ NextPathway
Building AI-powered automation tools to accelerate cloud migration of legacy SQL and ETL systems, including column-level lineage extraction across heterogeneous pipeline files.
Latest Posts
-
Agent很强，但我们的工作真的被加速了吗？
Agent生成得越来越快，但分散的Context和有限的人类review能力，正在成为团队真正提效的新瓶颈。
-
我在大厂做startup - 从执行型PM到探索型PM
在大厂内部做类似 startup 的探索，从目标明确的执行型 PM 转向寻找方向、快速验证和说服用户的探索型 PM。
-
From Demo to Dependable
A field guide to building AI product evaluations that survive real users, messy inputs, and the distance between a promising demo and a dependable product.