Threat Hunting Pipeline
Cybersecurity · Dify · SOC
False-negative detection for SOC log filtering with a closed-loop rule-generation workflow on Dify — turning missed alerts into new detection rules automatically.
DifyvLLMQdrantPython
Read more AI Engineer•Agentic AI•RAG · Data Platforms
AI EngineerBecoming aTechnical PM
I'm Nguyễn Hải Lâm — AI Engineer at Viettel Cyber Security, shipping agentic systems for Cybersecurity (SOC / Threat Hunting), Legal AI and Software Testing. Now moving toward Technical PM to own outcomes end-to-end.
3+
Years in AI Engineering
3
Domains: SOC · Legal · QA
1M+
Records Processed
National AI
Olympiad 2025
Trusted by teams at
Viettel Cyber SecurityAIZHBLabĐấu ThầuPTIT IECHuaweiVTV3 OlympiaIEEE AccessViettel Cyber SecurityAIZHBLabĐấu ThầuPTIT IECHuaweiVTV3 OlympiaIEEE Access
What I Do
Crafting AI as software,
not as a parlor trick.
Four overlapping practices I bring to every engagement — from the first whiteboard sketch to the dashboard you check at 2am. Each one ships with evals, observability, and a runbook.
Evals includedCost-awareProduction-grade
01
Agentic AI Systems
Multi-agent platforms with LangGraph, FastMCP and tool use — engineered for real workflows, not demos.
Available
02
Security & Legal AI
SOC threat hunting pipelines and legal-domain RAG — with evals, guardrails and closed-loop improvement.
Available
03
Data Platforms
Kafka · Spark · DuckDB pipelines processing millions of records — clean architecture, calm at scale.
Available
04
Technical PM
Translating ambiguous problems into shipped systems — scope, stack, evals, timeline and stakeholder clarity.
Available
About Me
Nguyễn Hải Lâm — AI Engineer at Viettel Cyber Security, based in Hanoi. Working across Cybersecurity (SOC / Threat Hunting), Legal AI and Software Testing. IEEE Access 2025 co-first author and National AI Olympiad 2025 winner.
3+ years across 5 organizations shipping agentic platforms, multi-tenant RAG, and data infrastructure — from research to product to production. Now transitioning toward Technical PM to own scope, delivery and outcomes across cross-functional teams.
Cybersecurity / SOCLegal AISoftware TestingAgentic AILLM EngineeringData PlatformsTechnical PMResearch → Product
Experience
- Viettel Cyber SecurityAI Engineer — Threat Hunting Pipeline, SOC log filtering, closed-loop rule generation2026 – Present
- AIZAI Engineer / Project Manager — Enterprise multi-agent platform (LangGraph · FastMCP)2025 – 2026
- HBLabJunior LLM Engineer — Software testing agents, test generation2025
- Đấu Thầu EcosystemJunior AI Engineer — Legal AI, hybrid search, multi-agent RAG, MCP2024 – 2025
- PTIT IECLLM Engineer — Research, AI teaching assistant, 200+ students mentored2023 – 2025
How I Work
A calm, repeatable process.
Same loop, every project. It bends with the problem but never breaks the discipline.
01
Discover
Understand the user, the data, and the unfair edge AI should create.
02
Design
Sketch the system — prompts, retrieval, evals, latency budget and failure modes.
03
Build
Ship a thin vertical slice fast, then harden with tests, logs and dashboards.
04
Iterate
Measure with real users, kill what doesn't work, double down on what does.
Proficiency
Where I'm sharpest.
Agentic AI · LangGraph · FastMCP95%
LLM Engineering & Evals (Dify · vLLM)92%
RAG & Vector Search (Qdrant)90%
Data Engineering (Kafka · Spark · DuckDB)88%
Python / FastAPI Backend90%
Technical PM & Delivery82%
12+
RAG systems
8
Production agents
99.9%
Service uptime
Featured Projects
Enterprise Multi-Agent Platform
Agentic AI · LangGraph · MCP
Production multi-agent platform with tool calling, inter-agent communication and multi-tenant orchestration for enterprise workflows.
LangGraphLangChainFastMCPFastAPI
Read more 
GitHub Analyzer
Data Engineering · 1M+ records
Large-scale data pipeline analyzing GitHub activity — streaming ingestion, lakehouse storage and fast analytical queries over millions of records.
KafkaSparkDuckDBMinIO
Read more 
Legal AI — Hybrid RAG
Legal AI · Multi-Agent RAG
Hybrid search + multi-agent RAG over Vietnamese procurement / legal corpora, with MCP tools for citation-grounded answers.
LangChainQdrantFastMCPPostgreSQL
Read more 
TestGen — QA Agent
Software Testing · Agentic AI
Agent that reads PRD/SRS and auto-generates test cases — cutting QA time and standardizing the test process across teams.
LangGraphSeleniumAppiumPython
Read more IEEE Access 2025 — LLM Prompting
Research · Co-First Author
Peer-reviewed study on LLM prompting strategies, published in IEEE Access (2025). Co-first author.
LLMPromptingEvaluationResearch
Read more Tech Stack
Agents / LLM
LangGraphLangChainFastMCPDifyvLLM
Retrieval
QdrantHybrid SearchReranking
Data
KafkaSparkDuckDBPostgreSQLMinIO
Backend
FastAPIPythonDocker
Observability
LangfusePrometheusGrafana
Research
IEEE Access 2025LLM PromptingAgentic AI
Awards & Achievements
National AI Olympiad 2025
Hội Tin học Việt Nam
Posts and Telecommunications Institute of Technology
Road To Olympia 21 — VTV3
Vietnam Television
Đường Lên Đỉnh Olympia, season 21 contestant
IEEE Access 2025 — Co-First Author
IEEE
Peer-reviewed paper on LLM prompting
Top 10 — Huawei Seeds For The Future
Huawei
Selected from a national applicant pool
Engineering Philosophy
- Treat AI as software engineering, not magic.
- Build for real users in production, not for demos.
- Prefer simple systems that scale over complex hype.
- Optimize for reliability, observability, and impact.
- Bridge engineering and product — ship outcomes, not features.
Kind Words
What collaborators say.
"Hải Lâm thinks about AI the way good engineers think about distributed systems — evals, latency, observability, cost. Calm execution, every time."
SE
Senior Engineering Lead
Viettel Cyber Security
"He owned our multi-agent platform end-to-end. LangGraph, FastMCP, multi-tenant — production-grade from day one."
PD
Product Director
AIZ
"Rare combination: research depth, product taste, and the discipline to ship. Our QA agent shipped in weeks, not quarters."
TL
Tech Lead
HBLab
Writing
Notes from the workshop.
Field notes on building AI systems that survive contact with real users — from retrieval edge cases to agent failure modes.
Read all essays01
Security May 2026· 11 min read
Closing the loop: turning SOC false-negatives into new detection rules with Dify
02
Agents Mar 2026· 9 min read
Designing multi-agent systems with LangGraph + FastMCP that survive production
03
Data Jan 2026· 7 min read
Kafka → Spark → DuckDB: a pragmatic lakehouse for 1M+ records
Frequently asked
What kind of engagements do you take on?
Agentic AI platforms, security/legal AI systems, software-testing agents, and the data infrastructure behind them. Greenfield builds or rescuing prototypes that need to ship.
Why the move from AI Engineer to Technical PM?
I've owned systems end-to-end across 5 organizations. Moving into Technical PM lets me combine that engineering depth with delivery, stakeholder alignment, and outcome ownership.
What's your default stack?
LangGraph + FastMCP + Dify for agents, Qdrant for retrieval, vLLM for serving, FastAPI on the backend, Kafka / Spark / DuckDB on data, Langfuse for observability.
Can clients trust you with production AI?
Yes — production systems at Viettel Cyber Security and AIZ, 1M+ records processed, IEEE Access 2025 co-first author, and a National AI Olympiad 2025 win.