Top 20 nationwide, grant up to USD 15,000
RE:ACT Youth Innovation Challenge 2026, UNDP Viet Nam & National Innovation Center, with Samsung
GreenLoop, selected for the Living Lab pilot · 2026
hi, i'm
Trần Đăng Khôi
AI & Systems Engineer · Ho Chi Minh City
I build AI systems that run on real hardware. AIoT engineering student at IUH, working from Verilog RTL and Jetson deployment up through multi-agent pipelines and the product side around them.
currently
- Leading GreenLoop (live): Top 20 nationwide at UNDP’s RE:ACT, with a pilot grant of up to $15K
- Building GuardianCo-Pilot for FPT’s Automotive Hackathon: Round 2, <100 ms on-vehicle fast path
- Two glioma MRI papers: a public preprint, and an IEEE ISBI 2027 submission as 2nd author
- Selected for the Aspire Leaders Program, Aspire Institute, Harvard Business School
- Open to junior engineering roles and research collaborations
it! 222
certs
awards
Awards & recognition
Eleven results across national and international competitions in 2025 and 2026, from a Tsinghua final in Beijing to a UNDP-backed pilot grant.
Finalist, team lead
Tsinghua Global SDG Case Competition 2026, School of Public Policy and Management, Tsinghua University
GreenLoop policy case “Trust Before the Leap”, Final Round in Beijing · Nov 2026
Top 20 of ~300 teams, Round 2
Automotive Hackathon 2026, FPT Software, open to professional engineers
GuardianCo-Pilot · 2026
Top 16 of 900+ teams, quarter-finalist
The Supply Chain Champion 2026, IBC · UEH Mekong
AI for sustainable supply chains in the Mekong Delta · 2026
Top 20 of 260 teams, solo entry
Swin Hackathon 2026, fintech AI track
AgentFlow Finance Guard · 2026
Top 30 of 439 teams
Asian Hackathon for Green Future 2026, ~1,500 participants from 22 countries, final at VinUniversity
GreenLoop, water & climate-resilient agriculture track · 2026
Passed national screening
Smart City IC Design Competition, 3rd edition, HCMC High-Tech Park Authority
AMBA AHB DMA IP core, team lead · 2025 to 2026
Round 2, top ~20 teams
RMIT Research Challenge 2026, equity research
BUY report on VFG, team BON QUA TAO · 2026
Second prize
Fusion Food Concepts Challenge, RMIT Food Technology & Nutrition Day 2026
2026
Top 14
Student Startup Idea Competition 2026, Banking Academy of Vietnam
GreenLoop · 2026
Top 90, global field
Wege Prize 2026, global circular-economy design competition, Ferris State University (USA)
GreenLoop · 2026
experience
Founding, leading, mentoring
The whole run in order: what I founded, what I led, what I competed in, and the teams I mentor now.
2026
Tsinghua Global SDG Case Competition 2026
Finalist, team lead · team WANDERERS
Wrote the English-language public-policy case “Trust Before the Leap”: should a Mekong Delta rice cooperative join Vietnam’s national low-emission rice program? Built on policy documents and a field survey, and invited to the Final Round in Beijing on Nov 1, 2026.
RE:ACT Youth Innovation Challenge 2026
Project manager · GreenLoop
Top 20 nationwide among young innovators aged 18 to 35, run by UNDP Viet Nam and the National Innovation Center with Samsung. GreenLoop won a Living Lab pilot slot with a non-refundable grant of up to USD 15,000 and six months of technical mentoring.
see GreenLoop open the live siteAutomotive Hackathon 2026, FPT Software
Project Manager & Integration Lead · GuardianCo-Pilot
Leading a team building an on-vehicle driver-safety AI OS, and owning the integration myself: the shared world model and the sub-100ms fast path that ties perception, safety and personalization together over gRPC and CAN. Through to Round 2, top 20 of ~300 teams.
see GuardianCo-PilotThe Supply Chain Champion 2026, IBC · UEH Mekong
Quarter-finalist · Round 3 Case Challenge
Top 16 of 900+ teams with a case on AI for sustainable supply chains in the Mekong Delta.
TRACE-Seg3D: Counterfactual Context Auditing for Robust 3D Glioma Segmentation under Institutional Shift
2nd author on the IEEE ISBI 2027 submission · 3rd author on the TRACE-Seg3D preprint
A counterfactual context-auditing framework that keeps 3D brain-tumor segmentation reliable when the MRI comes from a different hospital or scanner. I own model optimization, explainability and the ablation studies. Best cross-domain Dice of 0.591 BraTS→UTSW against 0.567 for the strongest baseline, and up to 52.6% fewer enhancing-tumor false positives.
see the paper arXivAspire Leaders Program, Aspire Institute, Harvard Business School
Selected fellow · 2026 cohort
A global leadership program that began at Harvard Business School, founded by HBS Professors Tarun Khanna and Karim Lakhani. Selected into its international cohort of emerging leaders.
Startup Runway 2026, University of Economics Da Nang
Mentor · 158 teams from 34 universities
Mentoring a student startup team on technical architecture and product feasibility at one of Central Vietnam's largest annual student entrepreneurship competitions.
Tsinghua Global SDG Case Competition 2026
Mentor · SPPM, Tsinghua University
Mentoring a team building an English-language public-policy case on health-insurance (BHYT) access for ethnic-minority communities in Vietnam: policy gaps, stakeholder research, and reform options tied to the health-equity SDGs.
Graduate mentees abroad
Mentor · Sorbonne University and Technical University of Munich
Mentoring a computer science master’s student at Sorbonne University and a biology master’s student at TU Munich on programming, applying AI to their field, and writing research papers.
Asian Hackathon for Green Future 2026
Top 30 of 439 teams · GreenLoop
A pan-Asian hackathon with about 1,500 participants from 22 countries and territories. GreenLoop made the final round at VinUniversity, Hanoi, in the water resources and climate-resilient agriculture track.
see GreenLoopFusion Food Concepts Challenge 2026
Second Prize · RMIT Food Technology & Nutrition Day
Second prize in RMIT’s food-concept challenge, a step well outside engineering.
Student Startup Idea Competition 2026, Banking Academy of Vietnam
Top 14 · GreenLoop
GreenLoop pitched as a startup and finished in the top 14 at the Banking Academy of Vietnam competition.
see GreenLoopRMIT Research Challenge 2026
Team BON QUA TAO · Business Description author
Passed the Round 1 online test with the team and moved into Round 2 financial modelling and report writing on Vietnam Fumigation JSC. I wrote the business description section. Top ~20 teams.
see the VFG reportWege Prize 2026
Top 90 · GreenLoop
A global circular-economy design competition run by Ferris State University in Michigan, with teams from 30+ countries. GreenLoop reached the top 90.
see GreenLoopZESG Project, UEH
Business & Tech Lead · AGRIBRIDGE
Owned both sides at once: the three-tier architecture, and the Business Model Canvas, go-to-market plan and financial model behind it.
see AGRIBRIDGEDatathon 2026
Team BON QUA TAO · EDA and pipeline lead
Led exploratory analysis and pipeline design on noisy finance and retail tabular data, then trained and cross-validated the models and turned the output into visualizations a business reader could act on.
see the Datathon workCareID
Founder & Tech Lead · health-tech side project
Founded the platform, ran 30+ user interviews, architected the stack, and built the prototype and web demo before shelving it in 2026.
see the CareID buildSwin Hackathon 2026, fintech AI track
Solo entrant · AgentFlow Finance Guard
The team withdrew, so I carried it alone: multi-agent architecture, LangChain orchestration, FastAPI serving layer, integration and final submission. Top 20 of 260 teams.
see AgentFlow2025
Smart City IC Design Competition, 3rd edition
Team lead · group PHENK · AMBA AHB DMA IP core
Led five engineers through RTL design, functional simulation and synthesis with PPA optimization of a DMA core for air-quality IoT sensor nodes. Passed national screening.
see the DMA coreThe National Judicial College
Judicial education · a course built for the bench
The National Judicial College trains judges across the United States. Its course on best practices for substance use disorder cases in the courtroom is the legal side of the same problem I studied clinically at Penn Medicine.
Penn Medicine · Addiction Medicine
Continuing Medical Education, University of Pennsylvania
A clinician-level program from Penn Medicine on substance use disorders in hospital and outpatient care: how they are recognized, how they are treated with evidence-based, patient-centered methods, and how care is coordinated across a multidisciplinary team.
Metabolic Health ML
Co-author · with an Oxford PhD researcher
Built the end-to-end pipeline on NHANES: clinical feature engineering, Random Forest classifiers with SHAP explainability, and validation of the TyG index and TG/HDL ratio as metabolic risk markers. The findings feed a paper aimed at peer review.
see the researchSmulie EdTech
Intern · product tester & contributor
Product testing and iterative feedback for an early-stage EdTech startup.
Stanford University School of Medicine
Continuing Medical Education · 63 CME activities completed
Worked through the continuing education Stanford Medicine runs for practicing clinicians: 60+ episodes of the Stanford Medcast series, from CAR-T outcomes and CRISPR therapies to cardiac amyloidosis and ChatGPT in medical education, plus Essentials of Palliative Care and AI Series: Introduction to Healthcare. It is the clinical grounding behind my medical-imaging research and health-tech work.
Cognita Education
Founder & Director · 7 subjects
Built and ran a tutoring centre end to end with no outside funding: recruitment, scheduling, quality management, curriculum alignment and a teaching partnership, over five months of operation.
PPE Violation Detection, computer vision course
Team Lead · YOLOv8 + TensorRT on Jetson Nano
Led the team for the computer vision course project: a real-time PPE compliance detector hitting 97.3% mAP@50 at 30 FPS on a Jetson Nano, with TensorRT quantization to fit the board’s power envelope.
see the PPE build2024 and earlier
Ho Chi Minh City University of Industry (IUH)
B.Eng, AIoT Engineering
Coursework in AI systems, embedded systems, digital IC design, microprocessors, IoT architecture and robotics.
Online free class organizer
Self-initiated
Co-organized an eight-month free online learning programme for middle-school students with no institutional backing: built the community, kept attendance up, and kept the sessions running.
learning archive
I learn fast, and I learn a lot
Most of what I build started as something I taught myself at night. 222 courses and certificates from 40+ institutions, including Stanford, Wharton, Cambridge, Johns Hopkins, IBM and Datawhale, across AI, medicine, finance and law. It is why I can move from Verilog to a clinical paper to an equity report without starting from zero. Search it or filter it.
222courses and certificates, 40+ institutions
222 showing
228 files came out of the folder. Seven are the same certificate saved twice, so 221 distinct ones are listed here, plus a piano programme that never went through any of these platforms. Competition results and programmes sit under Awards & milestones and are not counted in the 222.
selected work
Things I've built and shipped
The four I would walk you through first, with the architecture spelled out. Six more one click away.
Climate tech · project manager, business & tech
2026 to now
GreenLoop, circular agriculture with verifiable records
A web platform for the Mekong Delta and Central Highlands that turns rice straw, usually burned, into biochar and into evidence: every weighing slip, alert and farmer decision goes into a hash-chained ledger that buyers, certifiers and banks can check. Built with 29 data tables, ~100 APIs and 5 user roles, and it installs as an app on Android and iPhone straight from the browser, working offline for cooperative staff in the field.
problem
About 26 million tonnes of rice straw a year in the Mekong Delta, only ~30% collected and the rest burned; over 70% of farmers keep no digital records.
architecture
Field weighing + moisture → collection route → pyrolysis → biochar → hash-chained ledger → third-party verifier
my part
Ran the team across business and tech, built the explainable rule-based salinity alerts, and the unit economics: a 15% platform fee with sensitivity analysis and stop conditions, for a 6-month Living Lab pilot in Hòa Đê, Cần Thơ.
Top 20RE:ACT nationwide (UNDP)
$15Kpilot grant, up to
Top 30of 439 teams, Asia
Top 14 · Top 90BAV · Wege Prize
Open the live platforminstallable on Android & iPhone · demo accounts inside
Node/ExpressPythonInstallable PWA (Android, iOS)LeafletHash-chained ledger
Automotive AI · project manager, integration lead
2026 to now
GuardianCo-Pilot
An on-vehicle AI operating system that moves ADAS from reactive to predictive: a shared world model, a deterministic safety kernel, Vietnamese-language alerts the driver can understand, and XGBoost vehicle-health intelligence. Everything runs on the edge box, nothing round-trips to a server.
problem
Driver-safety systems react after the fact, and the warnings they give mean little to a Vietnamese driver.
architecture
Camera + CAN → perception → shared world model → safety kernel → Phi-3-mini alerts
my part
I own the shared world model and the fast path, and glue perception, decision and personalization together over gRPC and CAN into one demo that survives a live run.
<100msfast path
Top 20of ~300 teams
Round 2Automotive Hackathon 2026
Jetson Orin NXONNX RuntimeCANAndroid AutomotivePhi-3-minigRPCXGBoost
Multi-agent AI · solo
Jan to Apr 2026
AgentFlow Finance Guard
A multi-agent pipeline for real-time fraud detection at the business-logic layer, with LangChain orchestration behind a FastAPI serving layer.
architecture
Transactions → LangChain agents → business-logic fraud checks → FastAPI service
my part
All of it. The team withdrew mid-competition, so I designed, integrated and submitted the system alone, and still finished in the top 20 of 260.
Top 20of 260 teams
Soloentry
LangChainFastAPIPython
Computer vision on edge
2025
PPE Violation Detection
Real-time detection of missing protective equipment on a site camera, running entirely on a Jetson Nano. TensorRT quantization brought inference inside the board's power envelope without losing accuracy.
architecture
Camera → YOLOv8 → TensorRT engine → Jetson Nano → live violation alerts
97.3%mAP@50
30FPS
YOLOv8TensorRTCUDAJetson Nano
Digital IC design · team lead
Dec 2025 to now
AMBA AHB DMA IP Core
A DMA IP core for air-quality IoT sensor nodes, compliant with the AMBA AHB bus standard. I led five engineers through RTL design, functional simulation, and synthesis with PPA optimization. It passed national screening.
design
4-state FSM pipeline · double-buffer memory · 3 sensors packed into 32-bit words
5engineers led
PASSnational screening
Verilog RTLAMBA AHBModelSimCadence Innovus
AgriTech B2B · business & tech lead
2026
AGRIBRIDGE
A three-tier platform connecting agricultural by-product suppliers with green manufacturers: supply-demand matching, UUID batch traceability, role-based access, and a three-phase bottom-up financial model for a Mekong Delta pilot.
3-tierarchitecture
Breakevenmodelled
ReactFlaskSQL ServerBMC
HealthTech · founder, tech lead
2026 · shelved
CareID, a digital health passport
A portable health record for elderly patients and the families who look after them: HL7/FHIR records, wearable sync for HRV and SpO2, a Vietnamese medical OCR pipeline, an AI lifestyle coach, and an emergency QR profile that still works with no signal. Shelved in 2026; the user research and the architecture are what I took from it.
validation
30+ in-depth user interviews first, then 47 organic waitlist sign-ups in two weeks, before the product existed.
architecture
Wearable + scans → OCR (Vision API + Tesseract, human in the loop) → FHIR record → alert engine + RAG coach → offline QR
compliance
AES-256 encryption, HL7/FHIR conformance, and data localization under Decree 13/2023.
47sign-ups, 2 weeks
30+user interviews
2startup competitions
RAG/LLMHL7/FHIROCRAES-256Figma
Embedded AI
2025
Autonomous Robot Arm
YOLOv8 detection wired into 5-DOF arm control for the microprocessor course: a real-time spatial-reasoning pipeline that turns vision output straight into servo signals.
5DOF
Jetsonon device
Jetson NanoYOLOv8Python
Equity research
2026
VFG Equity Research
An equity research report on Vietnam Fumigation JSC (HOSE: VFG) with a BUY recommendation. I wrote the business description: company overview, the asset-light two-segment model, revenue and gross-profit segmentation, cost structure and FX exposure.
Round 2top ~20 teams
Equity researchDCF / P-EExcel
Data science
2026
Datathon 2026, team BON QUA TAO
Led EDA and pipeline design on noisy finance and retail tabular data, trained and cross-validated several models, and turned the outputs into visualizations a business reader could act on.
EDAlead
CVcross-validated models
Pythonscikit-learnEDA
research
Three papers in progress
Two on glioma MRI, a public preprint and an IEEE ISBI submission, plus a metabolic-health study. All with researchers abroad, all on real clinical datasets.
TRACE-Seg3D: Counterfactual Context Auditing for Robust 3D Glioma Segmentation under Institutional Shift
Third author · with a University of Melbourne PhD researcher · 2026 to now
A counterfactual context-auditing framework for 3D brain-tumor segmentation that holds up under scanner and institutional shift, built on a MedNeXt encoder-decoder and evaluated on BraTS and UTSW glioma MRI.
My part: model optimization, explainability integration, ablation studies, and co-authoring the paper.
PyTorchMedNeXtBraTSUTSW
read it on arXiv: 2607.07038
Counterfactual Context Auditing for Glioma Segmentation Failure Detection Across MRI Datasets
Second author · with Nguyen Linh Dan Le (University of Melbourne) and Nguyen Pham Hoang Le (UIT, VNU-HCM) · 2026
The 4-page companion to TRACE-Seg3D for the IEEE International Symposium on Biomedical Imaging. It turns the same counterfactual context audit into a failure detector: flagging the MRI scans where a glioma segmentation model is about to go wrong when it moves to a new dataset.
Failure detectionTrustworthy MLMRI
Metabolic Health ML
with an Oxford PhD researcher · 2025 to now
An end-to-end pipeline on the NHANES dataset studying metabolically unhealthy obesity: engineered clinical features, Random Forest classifiers with SHAP explainability, and validation of the TyG index and TG/HDL ratio as metabolic risk markers.
The findings feed a paper aimed at peer-reviewed publication.
scikit-learnSHAPNHANESRandom Forest
stack
I work across the stack
Silicon to edge AI to the business case. Each group lists the projects where I actually used it.
AI / ML
PyTorchscikit-learnXGBoostYOLOv8OpenCVSHAPLangChainMulti-agentRAG/LLMMedNeXtSentenceTransformers
used in AgentFlow · PPE · TRACE-Seg3D · Metabolic ML
Edge & systems
Jetson Nano / Orin NXTensorRTONNX RuntimeCUDACAN busgRPCIoT sensors
used in GuardianCo-Pilot · PPE · GreenLoop
Embedded & silicon
C (embedded)Verilog RTLAMBA AHBFSM designModelSimCadence InnovusPPA optimization
used in AMBA DMA core · Robot arm
Backend & web
PythonFastAPIFlaskNode/ExpressReactPWARESTSQL / SQL ServerLeaflet mapsDockerGitLinux CLI
used in GreenLoop · AgentFlow · AGRIBRIDGE
Product & business
Business Model CanvasCAC / LTV / breakevenDCF & P/E valuationGo-to-marketOperationsFigma
used in GreenLoop · AGRIBRIDGE · VFG report
Domains I have shipped in
Circular agriculture & biocharAutomotive ADASFintech fraudMedical imagingClinical data (NHANES, FHIR)Agri supply chainsEquity research
used in GreenLoop · GuardianCo-Pilot · AgentFlow · research
about
Engineer, and the person around the engineer
EducationB.Eng, AIoT Engineering · Ho Chi Minh City University of Industry (IUH)Business Administration, Innovation in Business · University of Economics, University of Đà Nẵng (DUE)Aspire Leaders Program · Aspire Institute, born at Harvard Business SchoolContinuing Medical Education, 63 activities · Stanford University School of MedicineContinuing Medical Education in addiction medicine · Penn Medicine, University of PennsylvaniaSubstance Use Disorder in the Courtroom · The National Judicial College
LeadingGreenLoop · project manager, business & tech · RE:ACT Top 20 (UNDP)GuardianCo-Pilot · project manager & integration lead · FPT Automotive Hackathon, Round 2Team WANDERERS · team lead · Tsinghua Global SDG Case Competition finalistAMBA AHB DMA IP core · team lead of 5 engineers · Smart City IC DesignAGRIBRIDGE · business & tech lead · ZESG Project, UEH
NameTrần Đăng Khôi
BasedHCMC, Vietnam
LanguagesVN native, EN working
FocusEdge AI & ML systems
StatusOpen to junior engineer roles
RepliesUsually within a day
“Find the gap, build the team, ship it, then fix what breaks.”
I'm an AIoT engineering student at IUH. Right now most of my time goes to GreenLoop, the climate-tech platform I lead that UNDP's RE:ACT picked for a funded Living Lab pilot, and to GuardianCo-Pilot, the on-vehicle safety OS my team took into Round 2 of FPT's Automotive Hackathon. Around that: AI systems, embedded work, digital IC design, two glioma papers, and mentoring teams at Startup Runway and Tsinghua.
What I like is the part where a model has to survive contact with hardware. Quantizing YOLOv8 until it fits a Jetson Nano's power budget. Getting a DMA state machine through synthesis. Keeping a fast path under 100 ms when CAN traffic and an LLM are fighting for the same box.
The business side matters to me for the same reason. On GreenLoop I built the explainable salinity alerts and also the unit economics, the pricing and the go-to-market plan, because a farm record nobody will pay to verify is just a database row. I built and ran Cognita Education end to end, and with CareID I learned to run 30+ user interviews before writing code, and to shelve an idea when it is time.
Resourceful mostly out of necessity: with no laptop around I'll SSH into the Jetson from a Galaxy A12 over Termux just to check on a training run.
Outside engineering I write and produce music. I composed and released the ballad “You Are My Mom” end to end with Suno and WavePad, and I run the whole content pipeline solo: scripts, thumbnails, editing, trend research. That's where the 19,000+ followers came from, on no ad spend.
Short version
- Leading 5 teams, from GreenLoop (UNDP-backed pilot, up to $15K) to GuardianCo-Pilot at FPT’s Automotive Hackathon
- 11 competition results, including a Tsinghua final in Beijing and top 20 of ~300 at FPT
- 2nd author on an IEEE ISBI 2027 submission; the TRACE-Seg3D preprint is public on arXiv
- Edge AI that runs for real: 97.3% mAP@50 at 30 FPS on a Jetson Nano, a <100 ms on-vehicle fast path
- Aspire Leaders Program, Aspire Institute, Harvard Business School
- Founded Cognita Education; mentoring 2 student teams and 2 master’s students abroad
- 222 courses from 40+ institutions, including Stanford and Penn Medicine CME
say hello
Open to engineering roles
Junior engineering roles, research collaborations and founding teams. I reply to everything, usually within a day.
direct
khoitran08102006@gmail.com
(+84) 786 397 687
facebook.com/khoi.tran.51632
Ho Chi Minh City, Vietnam