Sign In

Paul Lam

Running a Startup Horse Race: Business Models, Customer Discovery & Paid Proofs of Concept

Paul Lam is a seasoned entrepreneur and software engineer with 20 years of experience, known for his expertise in leading and transforming engineering teams within deep tech startups.

During his tenure as VP of Engineering at Motiva A.I., Paul demonstrated his proficiency in machine learning engineering by developing an enterprise-scale ML system that combined fine-tuned language models and reinforcement learning, catering to Fortune 100 clients.

An active open-source contributor, Paul has contributed to popular tools like the Cascalog data analysis library and created the widely-used docker-clojure image. He also developed the Stepwise library for software orchestration on AWS, benefiting genomics data pipelines at UW LPDx.

Paul's speaking engagements at conferences such as Strata London and the European Commission have covered a diverse range of topics, showcasing his expertise in functional programming, customer analytics, and system debugging.

With a educational background in MASc Biomedical Engineering from the University of Toronto and BASc Electrical Engineering from the University of Waterloo, Paul possesses a strong academic foundation supporting his extensive practical experience.

Throughout his career, Paul has held key roles in various organizations, from being a Co-founder & CTO at Spokepoint to serving as a Data Scientist at uSwitch.com and a Patent Examiner at the Canadian Intellectual Property Office, Industry Canada.

Highlights

Jun 1 · linkedin

CoALA acts as a blueprint for designing more capable and general-purpose LLM agents by outlining key components like memory structures, action spaces, and decision-making processes

May 29 · linkedin

Embodied AI agents are an exciting field, but the current focus might be missing their true potential.

Right now, the community is largely fixated on fully autonomous systems like devin.ai. However, the most valuable use cases could be in human-agent collaboration. Emerging frameworks such as Autogen and Crew AI are working on hierarchical finite state machine agents, but that’s just one approach in organizing these AI agents. Another interesting avenue is structural causal models, as discussed in this paper.

The way we design AI agents will significantly impact their ability to solve problems—think Conway's Law. For quick starts, frameworks like Autogen are helpful. But for building substantial products, you may need to develop your own solutions.

After reviewing the source code of a few AI agent frameworks, I noticed they often overlook established software engineering principles like the Actor model. As a result, they are duplicating efforts and missing out on proven efficiencies. Systems built on these frameworks may face operational issues like traceability and robustness.

Let's strive for smarter, more efficient designs that leverage the best of both human and AI capabilities.

Related Questions

What is Paul Lam's role at Stealth Startup?
How did Paul Lam transition from academia to a tech startup?
What projects has Paul Lam worked on at Stealth Startup?
What are Paul Lam's main responsibilities as a Full Stack Developer?
How has Paul Lam's experience at UBC influenced his career?
Paul Lam
Paul Lam, photo 1
Paul Lam, photo 2
Add to my network

Location

Greater Tokyo Area