What topics does this blog cover?

This site covers applied AI systems, LLM workflows, statistical ML, data engineering, data products, and experiment infrastructure with practical writing across AI/ML, product, and engineering.

How can I contact Yangming Li?

You can contact Yangming Li via email at liym1@hotmail.com, through LinkedIn, or by using the contact form on the website.

Yangming Li | Applied AI Systems for Real-World Workflows

Q: Who is Yangming Li?

Yangming Li is an AI Engineer and Product Builder based in Vancouver, focused on LLM systems, statistical ML, data engineering, data products, and experiment infrastructure for healthcare, finance, and enterprise teams.

Runtime Snapshot

class YangmingLi: @staticmethod def builds() -> list: return ["LLM systems", "Statistical ML", "Data engineering", "Data products", "Experiment infrastructure"] @staticmethod def serves() -> list: return ["Healthcare teams", "Finance teams", "Enterprise teams"] @staticmethod def outcomes() -> list: return ["Shipped workflows", "Faster decision loops", "Reusable internal tooling"] # Instantiate Yangming Li builder = YangmingLi() print(f"Builds: {builder.builds()}") print(f"Serves: {builder.serves()}") print(f"Outcomes: {builder.outcomes()}")

>focus_areas = 5

>industries_covered = 4

>delivery_mode = "prototype to production"

Evidence Strip

A faster read on trust, fit, and delivery

Built for teams that care about adoption, auditability, and workflow impact, not just model demos. The homepage now keeps the proof points visible and moves side quests into quieter corners.

Industries

Healthcare operations Finance and risk Public sector delivery Research environments

Capability Areas

LLM systems Statistical ML Data engineering Decision-support products

Delivery Modes

Dashboards and scorecards APIs and microservices Copilots and RAG workflows Monitoring-ready pipelines

Public Artifacts

Peer-reviewed publication Technical guides Selected portfolio work CFA and FRM credentials

Working Style

Builder mindset

Tool Builder Mindset

"I'm a tool builder. That's how I think of myself. I want to build really good tools that I know in my gut and my heart will be valuable. And then, whatever happens, is... you can't really predict exactly what will happen, but you can feel the direction that we're going. And that's about as close as you can get. Then you just stand back and get out of the way, and these things take on a life of their own."

Lab & Notes

Lower-priority paths live here

Study notes, essays, experiments, and certificates still matter, but they no longer lead the homepage story. They now sit behind softer entry points so the main narrative stays focused on work, proof, and outcomes.

Notes Essays Lab Certificates

This keeps the homepage pointed at collaborators and hiring managers first, while still giving curious visitors a place to explore the side material.

Selected Writing

Browse writing by topic. AI/ML, product, and engineering now live under one blog view.

【2026】 pytorch review01

[ML] A practical PyTorch review of tensors, shapes, broadcasting, reshaping, and torch.distributions.
【2025】[MLOP] Leveraging Docker in Machine Learning and Data Science

[MLOP] A comprehensive guide to using Docker for ML/DS projects - from development to deployment
【2025】[MLOP] A Must-Have Skill for Efficient Model Management and Deployment

[MLOP] Understanding MLOP principles and implementation with MLflow and Weights & Biases
【2025】[ML]Trustworthy machine learning.

[ML] What is trustworthy machine learning.
【2025】[ML]Fine Tune Large Language Model for sentiment analysis

[ML] sft Large Language Model
【2025】[ML]Fine Tune Large Language Model for sentiment analysis2

[ML] sft Tune Large Language Model2
【2025】[MLOP] Ray.

[MLOP]What is ray.
【2025】[ML]Why Have Decoder-Only Architectures Become the Standard in LLMs?

[ML] Understanding the dominance of decoder-only architectures in modern LLMs
【2025】[ML] Building an Enterprise-Level AI Agent for Document Transformation

[ML] A comprehensive guide to building document processing AI agents using LlamaReport and LlamaCloud
【2025】[ML] The Model Context Protocol (MCP): Building a Connected Future for AI

[ML] A comprehensive guide to understanding and implementing the Model Context Protocol for AI integration
【2025】[ML] Uncertainty Quantification for LLMs: Teaching with UQLM

[ML] A hands-on teaching guide for using UQLM to quantify and understand uncertainty in large language models.
【2025】[ML] Build Autonomous AI Workflows with n8n: A Technical Introduction to Agentic Automation

[ML] A comprehensive guide to building agentic automation systems using n8n's AI Agent nodes.
【2025】[Classical ML] Exploring Random Forest: A Powerful Ensemble Learning Algorithm

[Classical ML] Dive into the mechanics, advantages, and applications of Random Forest - an ensemble learning algorithm that combines multiple decision trees for robust classification and regression tasks.
【2025】[ML] Machine Unlearning: A Complete Technical Guide

[ML] A step-by-step guide to implementing machine unlearning systems with algorithms, APIs, and monitoring processes.

Yangming's Product Blog

Building a Product That Scales into a Company: Lessons from the 4U Framework

A comprehensive guide to scaling products into successful companies using the 4U Framework and other strategic approaches.
The Essence of a Successful Product: Insights for Product Managers

A deep dive into what truly makes a product successful from a product manager's perspective.
A Deep Dive into Jira: Agile Project Management for the Modern Team

A comprehensive guide to using Jira for agile project management and team collaboration.

Yangming's Engineering Blog

[Statistics] Choosing the Right Statistical Test for Survey Analysis

[BI] A comprehensive guide to statistical tests for analyzing survey data
[Data Engineering] Understanding Databricks: A Comprehensive Guide with Real-World Examples

[Data Engineering] A deep dive into Databricks features and implementation with practical examples
[Engineering] Understanding Kubernetes: A Comprehensive Guide

[Engineering] A deep dive into Kubernetes architecture, components, and practical implementation
[Engineering] Introduction to Polars: A Fast and Efficient Data Processing Library

A comprehensive guide to using Polars for high-performance data processing in Python
[Machine Learning] Deep Neural Networks (DNN) Explained: Core Principles and Implementation

A comprehensive guide to understanding deep neural networks (DNNs), including forward and backward propagation, optimization algorithms, and PyTorch implementation
[Engineering] Deep Learning Engineering with JAX: A Modern Approach

A comprehensive guide to using JAX for high-performance machine learning and numerical computing
[Engineering] Feature Flags as Experiment Infrastructure: A Practical Guide to A/B Testing at Scale

[Engineering] A deep dive into productionizing A/B testing—from 5 experiments/year to 50,000/year—covering CUPED variance reduction, sequential testing, Bayesian vs. frequentist methods, plus a complete SOP template and resource checklist.
[Engineering] From Zero to One: Building and Publishing a Python Package

[Engineering] A comprehensive guide to creating, testing, documenting, and publishing Python packages following modern best practices
[Statistics] Generalized Linear Models (GLM): A Comprehensive Overview

[Statistics] A deep dive into GLMs, their applications, and implementation in both statistical and machine learning contexts

Notes

Working notes, study artifacts, and lower-priority references that support the main body of work.

Carnegie Mellon University Advanced NLP Course Notes

These are my study notes from CMU's Advanced Natural Language Processing course. The notes cover fundamental concepts and advanced topics in NLP.
MIT Data Structure and Algorithms Course Notes

These are my study notes from MIT's Data Structure and Algorithms course. The notes cover fundamental algorithms, data structures, and their practical implementations.
MIT Principles of Computer Systems (6.826) Course Notes

These are my study notes from MIT's Principles of Computer Systems course. The notes cover distributed systems, concurrency, fault tolerance, and system design principles.
MIT Computation Structures (6.004) Course Notes

These are my study notes from MIT's Computation Structures course. The notes cover digital systems design, Boolean logic, computer architecture, and assembly language programming.

Selected Work

Representative work themes across healthcare, finance, and enterprise teams, centered on systems that improve real workflows instead of staying as demos.

LLM Systems

Applied AI for document and knowledge workflows

Designing LLM-assisted systems for document transformation, retrieval, review, and operational handoff in enterprise environments.

LlamaReport guide MCP guide n8n workflows

Data Products

Decision-support products for complex teams

Building analytics and product experiences that help healthcare, finance, and public-sector teams move from raw data to better operational decisions.

Product strategy Product fundamentals Workflow systems

Experiment Infrastructure

From experimentation to production delivery

Shaping the foundations for reproducible ML, faster iteration, and experiment-ready delivery with MLOps and engineering discipline.

A/B testing systems MLOps foundations Data platform guide

Essays & References

A quieter corner for essays, references, and ideas that inform how I build.

Why "Taste" Matters in Science — and in Technology

Exploring Nobel laureate Yang Zhenning's concept of 'taste' in research and how it applies to technology and product development. What separates the merely competent from the truly visionary in science and tech.
Interesting Resource: Calculating Empires

I recently discovered an fascinating interactive resource called "Calculating Empires: A Genealogy of Technology and Power Since 1500". This comprehensive visualization maps out the intricate relationships between technology, power, and human history over the past 500 years.
Knowledge Flow

An interactive platform for visualizing and exploring connected knowledge across various domains. Knowledge Flow helps discover relationships between concepts and ideas in a structured format.

Lab

A smaller corner for interactive prototypes and playful experiments. The slot machine stays here as a lightweight demo, not part of the main positioning story.

Risk Forecast Studio

Turn Monte Carlo into a finance or delivery risk story

Treat each run like a committee review: one path could be a quarterly portfolio outcome, another a launch program under delivery pressure. The upper chart shows how scenarios drift apart over time, and the histogram reveals where the ending cases really cluster.

10-90 risk band Sample cases Required hurdle

A balanced review setup where the hurdle still feels reachable, but the tail risk is visible enough to force a real decision.

Situation Quarter-close risk review

A lead is checking whether the plan still clears its hurdle before committing more capital, timeline, or scope.

Main Risk Variance can overpower the base case

The center line may look calm, but a few bad shocks widen the tail quickly and change the story for stakeholders.

Decision Lens Adjust before review day

Lower the hurdle, extend the horizon, or reduce exposure and scope if the success odds drift too low.

Scenario Paths

How the forecast can unfold

Preparing simulation...

Distribution

Where the ending scenarios cluster

Target not set

Beat Hurdle --

Expected Finish --

Median Case --

10-90 Band --

Baseline level 100

Opening portfolio value, buffer, or program progress before the uncertainty starts compounding.

Expected gain / velocity +0.45%

The average step change if the plan is working, before market shocks or delivery friction show up.

Shock / delivery noise 1.80%

Use this for market swings, scope creep, coordination drag, and other sources of variance.

Review horizon 72

More steps can help a plan recover, but they also give risk more time to compound.

Scenarios 900

More scenarios make the probability story steadier and the tails easier to trust.

Required hurdle 135

The level you need to hit for the review to be called a win, whether that means return, runway, or launch readiness.

The point is not to chase one perfect forecast. It is to see whether the plan still holds together once you let uncertainty show up honestly.

Decision Simulators

Three more playable tools for experiment design, AI economics, and representation learning

These are not generic calculators. One helps answer "can I trust this experiment read?", another stress-tests production AI economics, and the third turns high-dimensional structure into a neighborhood map you can actually interrogate.

Experiment Design Studio

A/B Test Power Simulator

Model the decision pressure behind a launch review: how much sample, how much noise, and how much real lift you need before a "winner" deserves trust.

Evidence Map

Null vs. uplift distribution

Calibrating read...

Runtime Pressure

How long you need to sit on the test

Timing pending

1 day 1 week 2 weeks 1 month+

Short tests feel faster, but they usually buy speed by borrowing confidence from the future.

Applied AI Economics

LLM Cost-Latency Simulator

Stress-test an AI workflow the way a platform lead would: traffic, context size, retries, and cache behavior all fight over the same latency and budget envelope.

Cost Stack

Where the monthly spend really goes

Illustrative economics

Latency Budget

How queueing and retries bend the p95

Health pending

Monthly Spend --

P95 Latency --

Failure Rate --

Cache Savings --

Model class

Average QPS 1.2

Steady state throughput across the month, not just the homepage demo spike.

Input context tokens 7,000

Prompt, system instructions, retrieved docs, and conversation history combined.

Output tokens 900

The average answer length or tool-augmented completion size.

Cache hit rate 35%

Shared prompts, reusable context, and semantic caching can shave cost without changing the UX.

Retry share 6%

The fraction of requests that need a retry because of timeouts, transient failures, or guardrail retries.

These numbers are illustrative economics, not live vendor quotes, but they are useful for understanding how quickly context size and retries turn into budget pressure.

Representation Learning Studio

UMAP / HDBSCAN Manifold Simulator

Compress a synthetic high-dimensional population into a neighborhood map and watch density structure survive, split, or dissolve as overlap, local scale, and minimum cluster size shift.

Embedding Surface

How local neighborhoods fold into 2D

Projection pending

Density Frontier

Where dense structure becomes noise

Cluster scan pending

Groups Found --

Noise Share --

Separation --

NN Trust --

Latent dimensions 18

The hidden feature space before the embedding compresses the geometry into a readable plane.

Seeded groups 4

How many underlying structures the synthetic population starts with before density blur and overlap kick in.

Overlap / noise 0.55x

Raise this to make manifolds smear together, inject border cases, and force the clustering logic to work harder.

Local neighborhood 12

Smaller values emphasize fine local detail; larger values smooth the embedding and favor broader structure.

Min cluster size 10

Acts like the patience of a density clusterer: tiny islands can disappear if they cannot support enough local evidence.

Manifold curl 0.85

Bends the hidden geometry so the map stops looking like clean Gaussian blobs and starts feeling more like real representation space.

Use this like a UMAP and density-clustering intuition board: local neighborhoods can survive the projection even while cluster boundaries become contestable.

Below it, the slot machine keeps the lighter arcade energy for visitors who want something more playful than probabilistic forecasting.

AI Slot Machine Demo

CREDITS

1000

WINS

BET

PAYLINES

MULTIPLIER

FREE SPINS

AI Pioneers & Payouts:

Geoffrey Hinton
"Godfather of AI"
x5 = 50

Yann LeCun
Facebook AI
x5 = 75

Andrej Karpathy
ML Educator
x5 = 100

Yoshua Bengio
Deep Learning Pioneer
x5 = 150

AI Research
Innovation
x5 = 200

Jensen Huang
NVIDIA CEO
WILD

Steve Jobs
BONUS

Contact

Email liym1@hotmail.com
LinkedIn Yangming Li
GitHub github.com/yml-blog
Google Scholar Scholar Profile
Location: Vancouver, BC, Canada
Book a Meeting

Pick a time directly on my calendar:

Prefer email? liym1@hotmail.com
Leave a Message

Recent Messages