Skip to main content

Vu Minh Hieu

Senior AI Engineer + Agentic AI, Hill Climbing, MCP, and Retrieval

Scroll to projectsScroll to projects

  • 95%

    of pull requests delivered through the AI SDLC

  • 8x

    LLM cost cut at held accuracy

  • 5M+

    users served by recommendations

  • 86,000+

    document sets processed a year

Kara: AI Meeting Agent

A real-time meeting-capture product across web, iOS and Android that acts on meetings end to end, from live capture to automated follow-up.

View project
Kara: app to agent service, tool interceptor, confirmation gate and MCP servers, with a trace-driven eval loopKara appWeb · iOS · AndroidBackend hubforwards chat, screen contextAgent service · FastAPI · SSEContext nodememory · knowledge indexExecute nodeReAct tool loopPostgrescheckpointsrewind unsafe turnsMemoryMem0 on QdrantRedis cacheTool interceptorinjects user & workspace scope · drops model IDshides injected params · opaque ID aliases (Redis)fail-closed on unknown IDs · egress filterUser confirmation gatewrites only · a human yesMCP serversInternal toolsmeetings · documentscontacts · to-dosConnectorGoogle Calendar · Drivestateless, token per callReal-estate CRMplatform MCPintegrationPartner bank CRMSalesforceintegrationBehind the internal toolsDocument searchdense + sparse, RRFon QdrantKnowledge graphper-user · to-do foldingFalkorDBLangfuse traces · eval loopAI Agent Evaluations: role-played scenarios vs baselineHill climbing: daily prod traces → draft merge request
Kara Voice: User, Receptionist, Backend and Expert layers, question down and answer upUserReceptionistBackendExpertKara appmic · speaker · screenGemini LiveVAD · acksnever answersdata itselfVoice sessionreceptionist promptanswer shapingVoice relay · shared turn runnerauth ticket · PCM16 relay · same chat turn as typed chatKara's existing agent:same data, numbers,refusals and safetyrules as typed chat.Main agent graphsame agent as typed chatMCP toolsread allowlistconsent for writes1 SPEAK9 HEAR + SEE2 ASK_KARA8 WHEN_IDLE3 TURN REQUEST7 TURN RESULT4 CHAT TURN6 ANSWER5 TOOL CALL

Kara Voice: A Realtime Voice Agent

A production voice layer for Kara where a realtime model is the receptionist and every data question runs the same agent turn as typed chat.

View project

Agentic SDLC: Spec-to-PR Delivery

An open-source Claude Code and Codex plugin that carries a change from approved spec to reviewed pull request, with a human gate at every handoff. My team at One Mount runs it daily.

View project
AI-native SDLC: six stages, one artifact each, a human gate after every stage1 · Planintent.mdinterview to set the intentHuman gate2 · Designspec.mdflagged concernsHuman gate3 · Buildplan.md, plan review, then codeengineer approves the planHuman gate4 · Testreal test outputevals on config changesHuman gate5 · Deployreviewer, then fresh verifierPR for human reviewHuman gate6 · Maintainbands.yaml, diagnosticiandrafts the next intent.mdHuman gateFast pathdeclared low-riskEnforced at thetool boundaryArtifact gateno write withouta committed planProd-guardblocks destructiveprod commands
TV360: offline and online pipelines to serving, recommendation and searchOffline · batchOnline · streamingBatch jobsSpark · Hadoop · HiveCandidates & embeddingstwo-tower modelcontent-based embeddingsNew titlescold start: candidatesfrom content embeddingsUser eventsviews and sessionsKafka · Flinkstream processingReal-time featuresnew users: session signalsfeature store in RedisRediscandidates · real-time features · serving cacheRecommendationsSearchJava / Spring serving12,000 TPS · app requests99.6% under 100 msOnline rankingcandidates from Redisranked on real-time featuresRules & popularitycomplement the ranking8 recommendation blocks5M+ activated usersLLM query understandingto structured searchElasticsearchBM25 full-textkNN vector searchPersonalized re-rank3 search blockstop search, autocomplete,resultsOnline A/B testingengagement 3–6x (recommendations) · CTR 2–4x (search)versus the default baseline

TV360 Recommendation & Search

Personalization for live TV, film and video-on-demand on Viettel's consumer streaming app.

View project

Contaminated Land MCP

An MCP server for contaminated-land site assessment on public Australian guidance (NEPM), built end to end in Python.

View project
Contaminated Land MCP: host to tools to redaction, retrieval and validationMCP hostany MCP-compatible hostMCP server · 4 read-only toolsCited searchpage-level citationsLab-result screeningdeterministic codeCriteria listingGuarded draftingresults sectionPII redactionclient names & addressesbefore text leaves the machineHybrid retrievalvector + full-text, rank fusionNEPM guidancepage-level indexDraft validatorchecks every citation and number

Snake Eyes at Midnight

A cute-but-dark boss-rush roguelite built in Godot: a tiny knight fights giant cursed toys.

View project

Hi there

Show me any AI system in the world and I can reverse-engineer its architecture for you: how it retrieves, reasons, calls tools, guards itself and what it costs to run. Then I can build it. Senior AI Engineer who designs, builds and ships LLM systems that other people use, end to end: retrieval with source citations, agentic workflows, tool use through MCP servers, and the evaluation, guard rails and cost work that keep them in production. Works directly with the people who use the tools: requirement sessions, demos and hands-on training for operations staff. Builds in Python every day. Previously AI Engineer then Technical Lead at Viettel (2021–2024), building recommendation and search for the TV360 streaming app.

Experience

  1. Jul 2025 – presentSenior AI Engineer, One Mount Group
  2. Q2 2026 – presentSenior AI Engineer, Techcombank (AI Transformation Center, concurrent)
  3. Jun 2025 – Sep 2025Senior AI Engineer, Vietnam Trade Promotion Agency (Ministry of Industry and Trade), remote contract
  4. Nov 2024 – Jun 2025Senior AI Engineer, BIDV
  5. Jan 2021 – Sep 2024AI Engineer → Technical Lead, Viettel Big Data Analytics Center

Certifications: NVIDIA-Certified Professional — Agentic AI (NCP-AAI); NVIDIA Certified Associate — AI in the Data Center; TensorFlow Developer Certificate; SQL (Advanced), HackerRank.

Awards: Top 1% internal company-wide performance rating and Key Talent selection, One Mount Group. Champion, One Mount Group internal AI Hackathon.

Send me a message
Vu Minh Hieu