Chin Keong Lam

Chin Keong Lam is the founder and chief executive of Patho.ai and an enterprise AI engineer at HP. He builds knowledge-augmented generation systems that combine structured expertise, specialized agents and quantitative reasoning to support complex business decisions.

Lam studied electronic and communication engineering at the University of Essex and worked in semiconductor engineering, electronic-design automation and precision instrumentation. He subsequently founded the solar-technology company Zolargus and Palo Alto Scientific, which applied machine learning to sports, fitness and recovery. Its profileMyRun project analyzed wearable-sensor data to guide runners; Lam’s patented runner-data processing system combines footwear-based pressure sensors with accelerometers.

At Patho.ai, Lam initially developed language-model applications for drug discovery, including the Llama-based Chem llama project, before expanding into expert advisory systems. His professional profile traces this progression from physical sensing and specialized hardware to enterprise AI.

How he designs expert AI systems

• Knowledge graphs that capture judgment. Lam organizes factual knowledge, previous experience, emerging insights, current circumstances and feedback into interconnected structures. A competitive-analysis system can consequently combine market data, historical campaigns, industry intelligence and changing business conditions to generate strategic recommendations.

• Quantitative reasoning beyond vector retrieval. Semantic search can retrieve relevant passages while missing relationships, multistep dependencies and precise calculations. Lam uses structured graph queries and function calling to produce evidence-based financial comparisons and other numerical answers, as demonstrated in his AI Engineer presentation.

• Multi-agent orchestration around shared memory. A supervisory agent coordinates specialists handling tasks such as market research and sentiment analysis, each updating a shared knowledge graph. Lam prototypes these workflows with n8n and accommodates OpenAI, Anthropic and on-premises models.

• Expert-guided graph construction. Automated extraction identifies candidate entities and relationships, while domain specialists refine the taxonomy and remove misleading connections. Lam recommends Neo4j graph-building tools and Docker as accessible implementation starting points.

His interest in deployment also extends to edge computing: in a public assessment of embedded-AI tooling, he described difficulties with Nvidia Jetson in hospital-oriented systems and explored AMD hardware and ROCm for robotics and multi-agent workloads.