Multi-agent orchestration
Agent teams with clear roles, tool boundaries, and hand-offs. Built with LangGraph, Bedrock AgentCore, and custom orchestration loops.
LangGraphAgentCoreMCP
AI Engineer · Security Engineer @ Crypto.com
Multi-agent AI
that ships real work.
I'm Vincent Ko. I build production agent systems, RAG pipelines, and detection automation — then measure whether they actually behave.
7x
backend speed increase
multi-threading + data structures
48%
faster agent responses
12 → 6.3 minutes
51%
lower multi-agent cost
$1.02 → $0.50 per run
9.25x
smaller cross-agent payload
13,994 → 1,303 chars
00 / Cyber Security
Certified Junior Penetration Tester eJPT
Practical pentesting & security fundamentals - increasingly relevant as agents are badly instructed to do hacking.
01 / Focus
AI systems that hold up in production.
Where agentic AI meets security engineering: automated detection, multi-agent orchestration, retrieval, tool wiring, and evaluation.
Detection engineering
Detection-as-code workflows: Sumo Logic CSE rules, MITRE ATT&CK gap analysis, log-source wiring, alert triage, and noise reduction.
Sumo Logic CSETerraformATT&CK
Evaluation & observability
Tracing, task-outcome and behavior evaluation, fallback-policy testing, and tighter agent loops without quality loss.
Agent tracingBehavior evalLiteLLM
Production AI backends
MCP server implementation, third-party MCP tool wiring and access controls, plus FastAPI/NestJS services, RAG, queues, and AWS infrastructure.
MCPFastAPINestJSAWS
02 / Personal project
SAGE — my personal multi-agent platform.
A personal multi-agent assistant built for my own workflows. It plans, delegates, calls tools, and streams results through a web UI — and it powers the chat on this page.
Personal use
Websocket bot
SAGE is a multi-agent assistant I designed and built from scratch: a lead orchestrator decomposes a goal, routes work to specialist agents like Mage, and combines their output.
It runs the small websocket bot behind the chat bubble here, so visitors can ask it about my experience. Under the hood: orchestration, tool/MCP wiring, evaluation, and failure handling.
- Lead orchestrator + specialist agents
- Tool and MCP wiring, with access controls
- Streaming chat UI; schedules, notifications, analytics in progress
- Tracing, behavior evaluation, and self-improvement
- Local-model friendly (DGX Spark)
Why it matters: a real system I use daily, not a demo. Happy to walk through
the architecture in an interview.
03 / Experience
From blockchain monitoring to detection automation.
Two chapters at Crypto.com: first building large-scale blockchain security software, now bringing multi-agent automation into the Detection Engineering department.
Crypto.com
AI Engineer, Security Engineer
AI Engineer + Security Engineer in the Detection Engineering team, automating detection workflows.
Agentic AI & automation
- Multi-agent use case design & development
- Accurate multi-agent orchestration
- Production RAG and MCP wiring/development
- Agent security: access controls and write-tool guards
- Agent self-improvement
Evaluation & observability
- Agent tracing & observability
- Task-outcome and behavior evaluation
- Fallback policy and accurate tool-use checks
- Agent loop optimization
Detection engineering
- Sumo Logic CSE and Detection as Code (Terraform)
- Log-source wiring and alert triage
- MITRE ATT&CK gap analysis
- False-positive / noise reduction
- CSE rule fine-tuning
Crypto.com
Security Engineer
Blockchain monitoring and multi-agent security review. Cut blockchain project development time from 2–3 weeks to 1 week.
AI Code Reviewer (multi-agent)
- Production multi-agent review system on AWS EC2 with FastAPI and producer-consumer
- LangGraph workflows for file indexing (PGvector) and deep analysis
- Tools for multi-repo referencing, symbol retrieval, RAG, and delegation
- Interactive GitHub reviewer + summarizer agent
- 100+ vulnerabilities found; 7x speed increase; 40% token reduction
Blockchain security software
- Large-scale blockchain monitoring (NestJS, microservices, AWS)
- 1.2M+ monthly active users; deposits, withdrawals, and trading volume
- 100+ repos automated with Claude Code agents
- Double-spending detection; new chains in 1 day
- Prometheus, Grafana, PostgreSQL, Kubernetes, Docker Compose
Earlier
Software Developer Intern @ HKU (2023)
Freelance Web Creator (2020)
04 / Projects
Selected work.
Production systems and personal builds. Company work is described publicly; personal projects are open where possible. 30+ open-source projects on GitHub.
Personal project
01
SAGE
A personal multi-agent assistant with a lead orchestrator, specialist agents, tool/MCP wiring, and a streaming web UI. It also powers the chat on this page.
Multi-agentMCPWebSocket
TracingBehavior eval
Crypto.com
02
AI Code Reviewer
Autonomous code review with LangChain and RAG, backed by a producer-consumer queue. Caught 100+ critical vulnerabilities and was well-received by developers and software VPs.
LangChainPostgreSQLFastAPI
RAGConcurrency
Crypto.com
03
Multi-Agent Detection Automation
Multi-agent workflows for Detection Engineering: detection-as-code, Sumo Logic CSE rules, MITRE ATT&CK gap analysis, and noise reduction.
Sumo LogicTerraformATT&CK
AgentCore
Crypto.com
04
Blockchain Monitoring
Transaction monitoring microservices for 1.2M+ users. Achieved a 7x execution-speed increase through multi-threading and data-structure optimizations.
NestJSAWSPrometheus
RedisMicroservices
Open source
05
JishoNotes
Full-stack Japanese learning webapp with dictionary and flashcards. Java Spring, ElasticSearch, VueJS.
Java SpringElasticSearchVueJS
Open source
06
AdvancedWebServer
Java web server built from scratch with HTTP/2 and WebSockets.
JavaHTTP/2NIO
WebSockets
Open source
07
Protocol Clients from Scratch
Java Socks5 and SMTP clients built from scratch to understand the protocols at socket level.
JavaSocks5SMTP
Networking
05 / Contact
Let's build something that works.
Open to conversations about agentic AI, detection engineering, backend scaling, or the evaluation work that makes AI systems trustworthy.
Prefer a conversation? The SAGE bot in the corner can answer questions about my background.