Agentic AI · Multi-Agent Systems · LLM/RAG & MCP · MLOps/LLMOps · Observability & AI Governance
📍 Mississauga, Ontario, Canada · 🌐 Open to Principal / Staff AI Engineer · AI Architect
I turn regulated, high-stakes business problems into production AI systems — with the evaluation, observability, and governance for reliable enterprise operations.
| 14+ years AI/ML & data science | 23+ years engineering · ~7 industries | 25× Edge AI speed-up | 5 publications · 2025–26 | 200+ professionals upskilled |
|---|
PRIME: Production-Ready, Rapid, Intelligent, Measurable, and Efficient AI Engineering. My approach combines delivery speed with evaluation, cost control, human oversight, and reliable operations.
| Letter | Meaning | Professional value |
|---|---|---|
| P | Production-Ready | Reliable, secure, scalable enterprise AI |
| R | Rapid | Faster delivery and engineering execution |
| I | Intelligent | Agentic AI, research-driven innovation |
| M | Measurable | Quality, performance, trust, business outcomes |
| E | Efficient | Cost-effective architecture and operations |
In practice: evaluation and observability as release gates · guardrails and rollback · PromptOps and reusable agentic patterns · model routing and cost control · governance from design, not after the fact.
Runnable projects and architecture references. Signals below summarize repository-reported tests, evaluations, or capabilities; prototype results do not establish production performance.
| Project | What it demonstrates | Signal |
|---|---|---|
| AgentForge Enterprise AI Platform | Governed agent runtime: hybrid RAG, multi-provider model routing, tenant entitlements, human approvals, hash-chained audit, and cost controls | 269 tests · 20/20 golden evaluation cases |
| Agentic Ops Services Platform | Multi-tenant agentic AI foundation: MCP-governed API boundaries, ReAct/ReWOO orchestration, Python/FastAPI, Angular, graph and hierarchical retrieval, and AWS Terraform definitions | 132 backend tests · 93% line coverage · locally validated |
| Agentic Project Management Advisor | CrewAI + local Ollama agents for planning, estimation, resource allocation, and forecasting | 42% faster planning · 91% structured-output compliance |
| Agentic Usage Billing Engine | Event-driven, idempotent usage-based billing: ingest → rate → aggregate → reconcile → MCP; serverless on AWS | 291 gating tests · 100% line and branch coverage |
| CognitiveMirror-Jev-MCP | Cognitive middleware for personal and enterprise agents: 14 MCP tools, REST API, memory, guardrails, approval workflows | 62 unit tests · 100% statement and branch coverage |
| ConversaFlow AI | Tenant-aware conversational runtime: calibrated intent classification, explicit dialogue graphs, typed tools, cited knowledge retrieval, REST/WebSocket APIs, and a TypeScript SDK | 171 Python tests · 94.44% intent accuracy on synthetic held-out data |
| Enterprise AI Reference Architecture | Vendor-neutral architecture knowledge base: AI/ML lifecycle gates, reference designs, reusable patterns, security controls, standards, and architecture decision records | 24 lifecycles · 8 reference architectures · 13 pattern catalogs |
| Industrial Agent Intelligence | Evidence-grounded industrial agents: facility ontology, policy-checked tools, numeric grounding checks, human escalation, and deterministic behavioural evaluations | 119 tests · 22/22 behavioural evaluation cases |
| Med Foundation Models | Whole-body MRI research framework: self-supervised learning with MAE, SimCLR, DINO and JEPA; Vision Transformers, multimodal fusion, and downstream fine-tuning | 132 tests · self-supervised and multimodal research framework |
| MGIIA — Mobility Assistant | YOLOv8 + DeepLabV3 + Llama 3.2 Vision for hands-free road-scene understanding | 94.7% detection accuracy · 40% fewer false positives |
| Quant Pulse Processor | Python/NumPy quantum toolkit: statevector simulation, five canonical algorithms checked against exact references, and pulse-level transmon modelling | 250 tests · 15/15 acceptance checks |
| Sentry-AI | Offline cybersecurity signal-processing demo: telemetry replay, statistical baselines, noisy-OR risk fusion, hysteresis | Explainable, reproducible detection |
| TrustMed AI Fabric | Governed content and AI reference implementation: evidence-backed claims, approval and expiry controls, provenance, MCP tools, cited retrieval, and hash-chained audit | 96 tests · 30 golden evaluation cases |
Selected delivery outcomes and implementation checks. Test-gate results are identified separately from business outcomes.
| Domain | Work | Measured outcome |
|---|---|---|
| Healthcare AI | Springer Nature case study — Top-tier hospital ED capacity | 30% fewer LWBS · 20% faster door-to-doctor · 14% lower discharged-patient LOS · 2,500–3,000 added visits/year |
| Clinical Decision Intelligence | North American clinical organization — QEMIDP / Pulse ED — MIMIC-IV + CXR, LangGraph 6-stage pipeline, 5 guardrail layers, HITL, local-first Ollama | 100% evidence attachment (test gate) · 100% guardrail reason-code traceability (test gate) · 5 guardrail layers · 6-stage pipeline · HITL + audit trace |
| Legacy Modernization | Local manufacturing company — Legacy-Lens — deterministic LangGraph supervisor, 4 specialist agents, 4 MCP servers, RAG + citation gate, HITL | 41/41 implementation tests passed · 100% citation-gated answers · 8/8 MCP tools · 3 golden eval cases · end-to-end audit trace |
| Edge AI + Multi-Agent | Local SME — FIVIS vehicle inspection — LangGraph multi-agent orchestration, RAG/VectorDB, Rust/WASM, ONNX, NVIDIA Triton | 25× performance · 70–85% lower latency · 40–60% scalability · grounded, auditable decisions |
| Drug Discovery | Market leader of the pharmaceutical industry — VAE + GNN + RL molecular pipeline | 90% valid molecules · 67.2% quality lift · 14.7% over baseline |
| Disaster Response AI | One of the world's healthcare organizations — AI-driven disaster resource matching; Neo4j knowledge graph, DSPy workflows, entity mapping, web scraping, GCP + AWS Lambda | 80–90% faster matching (weeks → hours) · 50%+ coordination efficiency · 60–70% faster resource discovery · 45–55% less manual coordination |
| Agentic Ops | Leading wholesale chain — APAMA project management advisor — CrewAI, Ollama, FastAPI | 42% faster planning · 91% compliance · 88% forecast correlation |
| HR & Finance Analytics | Asia top-tier services — governed Azure/Power BI platform | 45–55% reporting visibility · 30–40% forecasting accuracy |
| AI Security | Japanese bank — SecureNet AI WatchTower — GenAI data-leak monitoring across text, PDF, image, audio, video; risk scoring, RBAC, dashboards | 100% conversation risk scoring · 100% of 5 data types monitored · 3-role RBAC · 4 dashboards · 0% source data retained |
| Data Management & Governance | Asia top-tier bank — Project Orion and CATA — real-time and batch data ingestion with CDC/SCD2; Hadoop/Spark/Yarn, Airflow, S3, Collibra, Flink, PySpark, AWS/GCP | 40–50% better ingestion scalability · 45–55% less manual ingestion · 35–45% better data quality · 30–40% faster lake/warehouse availability · 50%+ analytics readiness |
| Enterprise MLOps | Top-five financial institution — AIML Workbench / SWOOSH-FRIDAY — Azure ML, Kubeflow, MLflow, DVC, Jenkins, Kubernetes, TruEra | 90% extraction accuracy · 50% faster turnaround · 5+ deep-learning models · RAI-governed |
| Organizational Analytics | Big Four consulting firm — ONA, workforce optimization | 50%+ org visibility · 30–40% lower key-person risk |
| AML & Compliance | World top-tier financial institution — Real-time screening, entity matching, alert workflow | 30–40% fewer false positives · 35–45% better screening accuracy |
View publication record on Google Scholar
Published work
- Can AI Create Emergency-Department Capacity Without Building More Capacity? — Springer Nature, September 12, 2026 · case study + teaching note
- An MCP-Driven Multi-Agent Enterprise Marketplace Platform Architecture for Myanmar Secondhand and Local Fashion Merchants — International Journal of Computer Applications, August 20, 2026
- Dynamic Functional Connectivity Patterns in Resting-State EEG for Classifying Learning Strategies — International Journal of Computer Applications, December 18, 2025
- Local Pattern Descriptor-based EEG Classification During Mental Arithmetic Tasks: A Comparative Analysis of Machine Learning Models — International Journal of Computer Applications, November 18, 2025
- Intelligent Movement-Controlled Brain-Computer Interface System Based on EEG Signal — River Publishers book chapter, May 25, 2025
Submitted manuscript
- Signal-Processing Detection with Human-Guarded Containment: A Live Network Security Agent — IJICS, Inderscience · submitted October 21, 2026; under review
Projects in progress
- QuantEdge MultiAgents Intelligent Decision Platform
- Cyber Security Agent
- Trustworthy Agent Factory
- LLM Inference
Academic service: Project supervision; reviewer, Digital Health (July 2026–present)
🏅 Recognition
“Kyaw Kyaw is a rare talent in the AI space... he served as the architectural backbone for our GenAI initiatives. What makes Kyaw Kyaw stand out is his platform-agnostic approach... he designs scalable, production-ready Multi-Agent and LLM systems that are built to last. He bridges the gap between complex data science and robust LLMOps seamlessly.” — Koko Maung, Senior Technical Architect, Munich Re
“One of the most talented professionals I've encountered in the field of AI/ML, Generative AI, and Large Language Models (LLMs)... KK has a remarkable ability to create effective Proofs of Concept (POCs)... Beyond his technical prowess, KK excels in mentoring junior team members.” — Saikiran Kannan, Asia Lead — Data & AI Strategy
“He has a rare talent for breaking down highly complex technical concepts, tools, and methodologies into clear, easily digestible explanations... He is an outstanding educator who genuinely empowers his students to succeed.” — Thomas Lwin, Technical, ToshibaTec Singapore
| Agentic AI & GenAI | Production ML & MLOps | Governance & Trust |
|---|---|---|
| Multi-agent orchestration, RAG, MCP/A2A, PromptOps, evaluation, tool calling | Architecture, deployment, tracing, CI/CD, model routing, cost optimization | Responsible AI, explainability, audit trails, PII, regulated-industry controls |
💼 Experience — 23+ years across ~7 industries
| Role | Organization | Period |
|---|---|---|
| Principal AI Engineer (R&D) | Independent / freelance, Mississauga | Jan 2026 – present |
| AIMLOps Engineer (R&D) | RadComm Systems, Ontario (during graduate studies) | Jun – Oct 2025 |
| Lead AIML Engineer | Omdena (during graduate studies) | Jun 2024 – Jun 2025 |
| Senior Manager — Solution Architect, AIMLOps & Data Science | Capgemini, Singapore | Nov 2021 – May 2024 |
| Data Analytics & Science Manager — Human Capital | Keppel Corporation | Jan – Apr 2021 |
| Senior Data Scientist — Data & AI CoE | Deloitte Consulting | Jan 2018 – Nov 2020 |
| Enterprise Solution Software Engineer (R&D) — Telecom | MyRepublic | Aug 2016 – Jan 2017 |
| Enterprise Software Developer (R&D) — Banking & Insurance | Lexmark Enterprise Software | Oct 2012 – Feb 2016 |
| Signal Processing Engineer (R&D) — Biomedical | Nitto Denko Asia Technical Centre | May 2011 – Sep 2012 |
| Vision Software Engineer (R&D) — Biometrics | NEC Asia Pacific | May 2008 – Apr 2009 |
| Research Engineer — Data Storage | DSI–A*STAR | May 2006 – Apr 2008 |
| Design Engineer (R&D) — Communication & Power | Panasonic | Oct 2005 – Apr 2006 |
| Service & Process Engineer — Semiconductor | ASM | Nov 2004 – Apr 2006 |
| R&D Engineer (Assistant Manager) — Power & Communication | Duwon Electronics, Yangon | Aug 2001 – May 2003 |
Enterprise highlights: MLOps Architect for unstructured data at Global Top Bank · Data-Ingestion Engineering at Top Asia Bank · AI and GenAI platforms for a major Japanese bank and insurers · AWS Volunteer Management System for the Singapore government · Organizational Network Analytics and Workforce Optimization at Deloitte.
🎓 Education & credentials
| Credential | Institution | Detail |
|---|---|---|
| MEng, Automation & Smart Systems | McMaster University, Canada | 2024–2025 · GPA 3.94/4 |
| Specialist Diploma, DevOps/MLOps | Republic Polytechnic, Singapore | 2022–2023 · GPA 3.5/4 |
| Innovation & Design Thinking | National University of Singapore | 2023 |
| MSc, Signal Processing | Nanyang Technological University, Singapore | 2003–2005 |
| BE, Electronics | Yangon Technological University, Myanmar | 1994–2001 · GPA 4.91/5 · First Class Honours |
Certifications: AWS Certified Cloud Architect · NICF PMP · Dataiku / Databricks · Big Data Analytics with Hadoop & Spark · DevOps CI/CD with Jenkins, Ansible & Kubernetes
🧰 Toolbox
| Area | Technologies |
|---|---|
| AI/ML Techniques | Multi-agent orchestration · agentic RAG · prompt engineering · tool calling · semantic retrieval · model routing · fine-tuning · reinforcement learning · self-supervised learning · transformers · vision transformers · RNNs · CNNs · classification · regression · clustering · NLP |
| LLMs & Agent Frameworks | OpenAI/GPT · Claude · Gemini · Llama · Qwen · LangChain · LangGraph · Deep Agents · OpenRouter · Ollama · Model Context Protocol (MCP) |
| AI Governance & Evaluation | Role-based access control · permission gates · human approval workflows · adversarial review · guardrails · evidence-grounded responses · reproducible benchmarks · quality gates · audit trails |
| AI/ML Libraries & Platforms | TensorFlow · PyTorch · Keras · NLTK · OpenCV · ONNX Runtime · YOLO · DINO · Databricks · Dataiku |
| Programming Languages & Application Development | Python · Java · C# · C++ · R · Scala · PySpark · Rust · TypeScript · JavaScript · SQL · FastAPI · React · Next.js · REST APIs · WebSockets · WebAssembly |
| Data Engineering & Governance | Apache Kafka · Apache Flink · Informatica · Hive · Impala · event-driven pipelines · streaming analytics · data reconciliation · BigID · Collibra |
| Databases, Retrieval & Caching | PostgreSQL · pgvector · Qdrant · MySQL · MongoDB · Redis · vector search · exact, semantic, and prefix caching |
| Cloud & Infrastructure | AWS EC2 · S3 · Lambda · Aurora · Azure Virtual Machines · AKS · Azure Machine Learning · Google Cloud Storage · BigQuery · Colab · Docker · Kubernetes |
| Observability, Analytics & Cost Optimization | Prometheus · Grafana · Langfuse · MLflow · distributed tracing · cost ledgers · inference budget controls · Tableau · Power BI · Google Analytics |
| Development & Quality Tools | Git · GitHub Actions · Jenkins · pytest · Ruff · mypy · Postman · SoapUI · Visual Studio · PyCharm · CI/CD workflows across Linux, Windows, and macOS |
- Part-Time Lecturer: University of London · Murdoch University (Kaplan) · Metis · MAGES Institute · Lithan
- Upskilled 200+ professionals in AI/ML and data science development and deployment
- Project Advisor: National Science and Technology Development Agency (2022–Aug 2024)
- Mentor: AI Saturdays · DataKind · Omdena
- Community: DeepEval · Boon Lay Zone F Resident Network · Mahamuni Buddhist Society · Toastmasters International
Applied AI engineering: agentic systems, real-time voice and multimodal assistants, LLMOps/evaluation, and production patterns.
LinkedIn · Portfolio · Google Scholar · YouTube · Medium · Kaggle · Email
Build fast. Measure everything. Ship with guardrails.