Production-Ready
Reliable, secure, scalable enterprise AI
AI / ML / OPS ARCHITECT · MISSISSAUGA, CANADA
I design governed, measurable AI systems that turn complex data into decisions people can trust — from agentic workflows and RAG to enterprise platforms and real-time computer vision.
01 / ABOUT / THE PRIME PHILOSOPHY
Principal AI Engineer · Forward-Deployed AI Architect
I turn regulated, high-stakes business problems into production AI systems — with the evaluation, observability, and governance for reliable enterprise operations.
Mississauga, Ontario, Canada · Open to Principal / Staff AI Engineer and AI Architect opportunities
Reliable, secure, scalable enterprise AI
Faster delivery and engineering execution
Agentic AI, research-driven innovation
Quality, performance, trust, business outcomes
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.
01 / WHAT I DO
Multi-agent orchestration, RAG, MCP/A2A, PromptOps, evaluation, tool calling
02 / WHAT I DO
Architecture, deployment, tracing, CI/CD, model routing, cost optimization
03 / WHAT I DO
Responsible AI, explainability, audit trails, PII, regulated-industry controls
Career chronology from the GitHub profile and supplied resume. The resume download continues to use the original supplied resume.
Independent / freelance, Mississauga
Role and dates from the GitHub profile or supplied resume. Read the source profile ↗
RadComm Systems, Ontario (during graduate studies)
Role and dates from the GitHub profile or supplied resume. Read the source profile ↗
Omdena (during graduate studies)
Role and dates from the GitHub profile or supplied resume. Read the source profile ↗
Capgemini, Singapore
Role and dates from the GitHub profile or supplied resume. Read the source profile ↗
Keppel Corporation
Role and dates from the GitHub profile or supplied resume. Read the source profile ↗
Deloitte Consulting
Role and dates from the GitHub profile or supplied resume. Read the source profile ↗
Freelance
Role and dates from the GitHub profile or supplied resume. Read the source profile ↗
MyRepublic
Role and dates from the GitHub profile or supplied resume. Read the source profile ↗
Lexmark Enterprise Software
Role and dates from the GitHub profile or supplied resume. Read the source profile ↗
Nitto Denko Asia Technical Centre
Role and dates from the GitHub profile or supplied resume. Read the source profile ↗
NEC Asia Pacific
Role and dates from the GitHub profile or supplied resume. Read the source profile ↗
DSI–A*STAR
Role and dates from the GitHub profile or supplied resume. Read the source profile ↗
Panasonic
Role and dates from the GitHub profile or supplied resume. Read the source profile ↗
ASM
Role and dates from the GitHub profile or supplied resume. Read the source profile ↗
Duwon Electronics, Yangon
Role and dates from the GitHub profile or supplied resume. Read the source profile ↗
| 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 |
AWS Certified Cloud Architect · NICF PMP · Dataiku / Databricks · Big Data Analytics with Hadoop & Spark · DevOps CI/CD with Jenkins, Ansible & Kubernetes
| 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 |
02 / PUBLIC WORKS / RUNNABLE PROOF
Runnable projects and architecture references, with direct repository links and the scope behind each reported result.
| Project | What it demonstrates | Results & Validation |
|---|---|---|
| AgentForge Enterprise AI Platform | Governed agent runtime: hybrid RAG, multi-provider model routing, tenant entitlements, human approvals, hash-chained audit, and cost controls README scope: Synthetic-data reference implementation; 96% is combined branch-aware coverage. Production provider, storage and identity replacements are documented. | PROFILE-REPORTED96% combined coverage · 269 tests passed · 100% offline golden-case pass rate (20/20) |
| 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 README scope: README reports 132 backend tests and 93% line coverage; the profile reports 136 and 91%. Measurement chronology is unresolved. Terraform is not applied; production-scale testing remains pending. | PROFILE-REPORTED91% backend line coverage · 136 backend tests passed · local validation; production-scale testing pending |
| Agentic Project Management Advisor | CrewAI + local Ollama agents for planning, estimation, resource allocation, and forecasting README scope: Current standard-library core reports 142 tests and 100% line/branch coverage. CrewAI/Ollama remains an optional prototype outside CI; the historical planning/compliance percentages are not established by those core tests. | PROFILE-REPORTEDPreviously reported: 42% faster planning · 91% structured-output compliance; supporting local benchmark record not located |
| Agentic Usage Billing Engine Included in the XERJ Corpus Hub as a reference for coding agents and developers. View XERJ Corpus Hub ↗ | Event-driven, idempotent usage-based billing: ingest → rate → aggregate → reconcile → MCP; serverless on AWS README scope: Local proof of concept: 291 gating tests plus 3 optional performance tests. AWS is a target blueprint; the README says its cloud deployment is not yet deployable. | PROFILE-REPORTED100% line and branch coverage · 291 gating + 3 performance tests passed |
| CognitiveMirror-Jev-MCP Live on Agensi.io. View on Agensi.io ↗ | Cognitive middleware for personal and enterprise agents: 14 MCP tools, REST API, memory, guardrails, approval workflows | PROFILE-REPORTED100% statement and branch coverage · 62 unit tests passed · live PostgreSQL and container checks recorded |
| ConversaFlow AI | Tenant-aware conversational runtime: calibrated intent classification, explicit dialogue graphs, typed tools, cited knowledge retrieval, REST/WebSocket APIs, and a TypeScript SDK | PROFILE-REPORTED94.44% intent accuracy · 94.50% macro F1 (72 synthetic held-out examples) · 171 tests passed |
| Enterprise AI Reference Architecture | Vendor-neutral architecture knowledge base: AI/ML lifecycle gates, reference designs, reusable patterns, security controls, standards, and architecture decision records | PROFILE-REPORTED72 starter-kit smoke tests passed · 171 Markdown files and 62 diagrams checked; architecture guidance, no measured business-impact percentage |
| Industrial Agent Intelligence | Evidence-grounded industrial agents: facility ontology, policy-checked tools, numeric grounding checks, human escalation, and deterministic behavioural evaluations | PROFILE-REPORTED100% behavioural-case pass rate (22/22) · 79.94% combined coverage · 119 tests passed |
| 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 README scope: 132/132 is the reported test pass rate, not code coverage or clinical accuracy. | PROFILE-REPORTED100% recorded test pass rate (132/132); clinical accuracy and downstream model gains not established |
| MGIIA — Mobility Assistant | YOLOv8 + DeepLabV3 + Llama 3.2 Vision for hands-free road-scene understanding README scope: Current decision layer reports 109 tests and 100% core coverage. The original YOLO/DeepLab/Llama-Vision path is outside CI, with no established model accuracy, latency or safety certification. | PROFILE-REPORTEDPreviously reported: 94.7% detection accuracy · 40% fewer false positives; supporting local benchmark record not located |
| Quant Pulse Processor | Python/NumPy quantum toolkit: statevector simulation, five canonical algorithms checked against exact references, and pulse-level transmon modelling | PROFILE-REPORTED96% coverage · 100% acceptance pass rate (15/15) · 250 tests passed; simulator validation |
| Sentry-AI | Offline cybersecurity signal-processing demo: telemetry replay, statistical baselines, noisy-OR risk fusion, hysteresis README scope: Offline synthetic replay demo. The README does not enumerate a test count; the profile reports 7/7. Real-data precision and recall are not evaluated. | PROFILE-REPORTED100% recorded test pass rate (7/7) · two synthetic attack replays reach ALERT; real-data precision/recall not evaluated |
| 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 README scope: Deterministic synthetic reference implementation; reported golden-case results do not establish clinical efficacy or production deployment. | PROFILE-REPORTED100% deterministic golden-case pass rate (30/30), citation validity and refusal accuracy · 90% branch-aware coverage · 96 tests |
03 / PRIVATE WORKS / ENTERPRISE DELIVERY
Anonymized enterprise engagements from the public profile. Outcomes are self-reported; implementation test gates are identified separately.
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
Discuss this engagement ↗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
Discuss this engagement ↗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
Discuss this engagement ↗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
Discuss this engagement ↗Market leader of the pharmaceutical industry — VAE + GNN + RL molecular pipeline
90% valid molecules · 67.2% quality lift · 14.7% over baseline
Discuss this engagement ↗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
Discuss this engagement ↗Leading wholesale chain — APAMA project management advisor — CrewAI, Ollama, FastAPI
42% faster planning · 91% compliance · 88% forecast correlation
Historical profile figures; the supporting local benchmark record was not located. Current deterministic-core tests do not validate these percentages.
Discuss this engagement ↗Asia top-tier services — governed Azure/Power BI platform
45–55% reporting visibility · 30–40% forecasting accuracy
Discuss this engagement ↗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
Discuss this engagement ↗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
Discuss this engagement ↗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
Discuss this engagement ↗Big Four consulting firm — ONA, workforce optimization
50%+ org visibility · 30–40% lower key-person risk
Discuss this engagement ↗World top-tier financial institution — Real-time screening, entity matching, alert workflow
30–40% fewer false positives · 35–45% better screening accuracy
Discuss this engagement ↗Reported outcomes supplied by Kyaw; not independently benchmarked in this portfolio update.
Discuss this engagement ↗Private repositories and client documents are not published here. Read the source profile ↗
04 / OPEN-SOURCE WORKS / COMMUNITY
Upstream contributions, active proposals and ongoing community work. Fork creation and open pull requests are distinguished from merged contributions.
| Project | Contribution status |
|---|---|
| DeepEval | Three contributions by MrKyaw, verified 2026-11-16:
Proposed contributions; the listed pull requests have not been merged. |
| promptfoo | Contribution ongoing — fork created; continuing upstream PR or commit |
| vLLM | Contribution ongoing — fork created; continuing upstream PR or commit |
| REA | In Progress |
05 / RESEARCH & SOLUTIONS / PUBLICATIONS
*Springer Nature*, September 12, 2026 · case study + teaching note
Find publication on Google Scholar ↗*International Journal of Computer Applications*, August 20, 2026
Find publication on Google Scholar ↗*International Journal of Computer Applications*, December 18, 2025
Find publication on Google Scholar ↗*International Journal of Computer Applications*, November 18, 2025
Find publication on Google Scholar ↗*River Publishers* book chapter, May 25, 2025
Find publication on Google Scholar ↗IJICS, Inderscience · submitted October 21, 2026; under review
Find publication on Google Scholar ↗Ongoing paper with a US team member
Discuss this research ↗Public repository access is unavailable. View the profile reference ↗Reviewed paper with a Canadian team member
Discuss this research ↗Public repository access is unavailable. View the profile reference ↗Ongoing paper with an India team member
Discuss this research ↗Public repository access is unavailable. View the profile reference ↗Ongoing paper with a Japan team member
Discuss this research ↗Public repository access is unavailable. View the profile reference ↗Ongoing paper
Discuss this research ↗Public repository access is unavailable. View the profile reference ↗Academic service: Project Supervision · Research Reviewer, Digital Health (July 2026–present)
Architecture and hands-on delivery across AI, cloud, data and enterprise systems. Explore the career timeline and generate a resume for your role.
Clinical multi-agent systems, EEG classification, governed AI and signal-processing security.
Evaluate the product portfolio and reported outcomes. Contact Kyaw to discuss a specific problem or product opportunity.
06 / RECOGNITIONS / ARCHITECTURE · COLLABORATION · TEACHING
Perspectives on architecture, technical leadership, mentorship and teaching.
architectural backbone for our GenAI initiatives
Highlights architecture leadership across multi-agent systems, LLM applications and platform-independent enterprise delivery.
excels in mentoring junior team members
Recognizes practical AI proof-of-concept development and mentorship alongside AI/ML and generative AI expertise.
clear, easily digestible explanations
Highlights the ability to explain complex engineering concepts and support learners through effective teaching.
Short excerpts and summaries from recommendations reproduced in the GitHub profile. LinkedIn may require sign-in to read the originals. Read the source profile ↗
Read LinkedIn recommendations ↗THE OPERATING SYSTEM
Evidence-first retrieval, clear evaluation gates, and observable decision paths.
Human approval, permissions, audit trails, and responsible AI built in from day one.
Connect architecture to outcomes: reliability, cost, speed, quality, and adoption.
A RESUME THAT SPEAKS YOUR LANGUAGE
Enter the job posting link to check its requirements against Kyaw’s source resume. Organization name is optional. We select relevant experience and refine supported wording for the opportunity, then prepare your two-page PDF with original facts and employment dates preserved. You can contact Kyaw for further review.