Applied AI Systems
LLM applications, retrieval-augmented generation, agents, document intelligence, and AI system integration for practical workflows.
Yangming Li
About Yangming Li
Yangming Li is an Applied AI Engineer, Data Scientist, and AI Product Builder focused on LLM systems, evaluation, statistical machine learning, NLP, data products, and experiment infrastructure for useful production AI.
Expertise areas
Where I focus
These themes connect the technical articles, project notes, and product thinking across the site.
LLM Evaluation
Evaluation design, uncertainty checks, model behavior analysis, and review loops that make generated outputs more dependable.
Machine Learning & NLP
Statistical machine learning, NLP, topic modeling, sentiment analysis, and trustworthy ML methods for applied data problems.
Data Products
Analytics systems, decision-support products, data engineering, and product experiences that turn data into operational judgment.
AI Product Builder
Product strategy, launch trade-offs, workflow design, and the decisions that move AI from promising demo to adopted system.
System approach
How I approach AI systems
I treat AI systems as workflow products, not isolated model demos. The practical work is making the system understandable, testable, and useful when people depend on it.
- Retrieval designDesign context pipelines that expose the right evidence and make document-grounded answers easier to inspect.
- Schema validationUse structured contracts so generated outputs can be parsed, checked, routed, and corrected before they reach users.
- UncertaintySurface ambiguity, confidence gaps, and failure modes instead of pretending every response has equal reliability.
- Human reviewKeep expert judgment in the loop for high-impact decisions, escalation paths, and quality feedback.
- EvaluationMeasure behavior with task-specific tests, adversarial examples, regression checks, and production feedback.
- Production usefulnessConnect AI outputs to real workflows, adoption constraints, latency needs, auditability, and business value.
Selected writing
A few representative notes
A compact set of posts that show the range of the site: AI engineering, LLM evaluation, data products, and product-minded experimentation.
AI Agent Evaluation Is More Important Than Prompting
Why production AI agents need custom eval sets, trajectory checks, calibrated judges, regression tests, and business-ready metrics.
Read articleTesting and Evaluating Copilot Agents
Schema-first evaluation, test sets, release gates, and human review for Copilot-style agents.
Read articleA 90% Accurate Model That Still Loses Money
Why high-accuracy churn prediction can improve renewal rate while losing revenue, and how uplift thinking targets incremental renewals instead of risk scores.
Read articleBeyond A/B Testing: Uplift Modeling in Industry
A practical guide to CATE, meta-learners, Qini, AUUC, decile lift, online experiments, and incremental ROI.
Read articleDatabricks Lakehouse Guide
Delta Lake, MLflow, Unity Catalog, and data platform implementation notes.
Read article
External profiles
Find more context
Use these profiles for professional background, code, and implementation context.