LLM systems
RAG, LangGraph, LangChain, MCP, tool-calling agents, SEA-LION, prompt management, human-in-the-loop workflows.
Senior Software Development Engineer (AI Agentic) at AMDSenior SDE (AI Agentic) at AMD
Haoming Koo
Making AI systems you can rely on.
Now building agentic AI at AMD. I graduated from AI Singapore as a Certified AI Engineer after seven years building production systems at Micron.
7yrsMicron Technology
$600M+Program impact
8Public systems
How I build
I build AI systems the same way, from problem to production.
A clear problem statement, a design that survives scale, a working baseline, rigorous evaluation, and a deployment that keeps running after launch.
01
Clarify the problem
What decision this needs to support, and what success actually looks like.
02
Design for scale
Data flow, latency budget, and where the system has to bend as load grows.
03
Build the baseline
The simplest version that could plausibly work, shipped first.
04
Evaluate and optimize
Ground-truth sets, error analysis, and tuning against the metric that matters.
05
Ship with CI/CD
Tests, staged rollout, and monitoring so it keeps working after launch.
Core stack
Applied AI, from workflow to deployment.
The work spans the parts recruiters now screen for: LLM systems, retrieval, agents, evals, observability, ML models, APIs, data pipelines, and shipped user interfaces.
Evals and observability
Evaluation harnesses, Langfuse, Arize Phoenix, MLflow, trace review, failure analysis, source grounding.
ML and computer vision
LightGBM, scikit-learn, PyTorch, TensorFlow, YOLOv8, OpenCV, MediaPipe, SHAP, feature engineering.
Product engineering
FastAPI, React, TypeScript, Dash, SQLAlchemy, PostgreSQL, SQLite, auth, rate limits, background jobs.
Data and deployment
Docker, Railway, GitHub Actions, Playwright, APScheduler, GCP, AWS, Snowflake, Airflow, Kedro.
Enterprise delivery
Workflow automation, semantic search, text-to-SQL patterns, root-cause investigation, stakeholder reviews, cross-site rollout.
Selected work
Real products, shown through the interfaces people actually use.
01 / 04Search 72,000+ Singapore roles
01 · Job search
Job Hunter SG
Searches Singapore roles, compares them with a resume, and guides users through evidence-bound resume building and edits without inventing experience.
- Role: Product, AI workflow, backend, and frontend
- Proof: Search, matching, guided resume chat, agent review, and export
- Stack: React, FastAPI, SEA-LION, sentence-transformers
01 / 03Explore seasonal conditions across Japan
02 · Travel intelligence
Japan in Seasons
Turns seasonal forecasts into a searchable map, public API, and MCP server that assistants can query through 17 read-only tools.
- Role: Data product, MCP interface, deployment, and discovery
- Proof: 1,700+ places with live seasonal and weather sources
- Stack: TypeScript, MCP SDK, Node HTTP, Zod
01 / 04Interrogate price structure with visible evidence
03 · Market intelligence
Trader Koo
Turns market data into source-dated research, ranked setups, calibrated evidence, and explicitly simulated paper-trade learning.
- Role: Research pipeline, model evidence, risk controls, and product interface
- Proof: Source freshness, archived calibration, critic gates, and paper-only audit
- Stack: React 19, FastAPI, Python, SQLite
More live products
LionWeather
Combines live Singapore weather feeds, radar, saved locations, and a temporally validated rainfall model in one operational view.
Amex Explorer
Map-first Singapore benefit explorer with official source hashes, scheduled refreshes, and stale-source alerts.
Photo Compliance Studio
Checks passport and ID photos with face landmarks, segmentation, crop guidance, and country rules.
Preflight
Triages CSV and Parquet datasets through schema checks, EDA, baseline models, and validation metrics.
MinMax Wine
Compares Singapore wine offers using normalized bundle prices, identity matching, confidence checks, and deal scores.
About
2018 to 2025
Micron Technology
Led AI-enabled transformation across four fabs (Singapore, Boise, Hiroshima, Taichung), with a team of 6 direct and up to 9 indirect engineers and a $9M portfolio. Co-led low-power DRAM and HBM yield programs, described here without customer or internal programme details. Built ML risk-scoring that lifted yield 1–2% at production scale, a GenAI SOP workflow scaled to ~3,000 engineers, and a digital twin that cut chemical use 60%. ResNet-50 wafer-defect classifier (Innovation Award). Work identified $600M+ in opportunities, with $50M+ realized.
Jan to Jul 2026
AI Singapore · AIAP
Graduated as a Certified AI Engineer after shipping production ML work across the 100E programme, vision, sequence modelling, RAG, evals, and MLOps.
2026 to now
AMD
Senior Software Development Engineer (AI Agentic). Building end-to-end agents with tool use, memory, function calling, structured outputs, RAG, context engineering, evals, and production integrations.
Always
Building AI
I keep side projects practical: real URLs, real data where possible, validation gates, and clear boundaries when something is demo or research-only.
Recent writing
View all ›From 73,857 Long COVID Records to One Plausible Treatment Lane
An auditable PubMed pipeline, six randomized rehabilitation trials, and a careful explanation of what REGAIN actually supports.
Read ›A Visual Guide to Simulating a TSV from Etch to Polish
Building a TSV process simulation with ViennaPS: process handoffs, tunable inputs, failure modes, measurements, and model limits.
Read ›I Am Now a Claude Certified Architect. Here's What I Wish I Knew First.
What the Anthropic certification family covers, why I chose this track, what to prepare, and the paced game plan that got me there.
Read ›Need AI systems people can trust and use?
Senior Software Development Engineer (AI Agentic) at AMD, based in Singapore.
[email protected]