Sang Truong · Senior AI Engineer · Applied AI Systems

I build the models, data and software that make AI useful.

I lead AI engineering at EPIC TECHNOLOGY and still work directly on architecture and code. Recent work includes credit-risk ML on AWS, clinical AI for hospital physicians, production video intelligence and factory-floor inspection. Before EPIC, I built core enterprise-agent orchestration at FPT Software; my research and independent quant work inform how I evaluate models and build reliable systems.

Ho Chi Minh City, Vietnam · GitHub @sangtrx

View flagship systems Technical depth Résumé

Current roleHead of Artificial IntelligenceAI / software engineering8+ yearsGraduate degreeMS Computer Engineering · 4.0/4.0ResearchIJCV · AAAI Oral · IEEE JBHI · Poultry Science

Technical depth

One engineer. Four deep system domains.

The headline stays simple. The depth comes from owning the hard boundaries: model versus truth, research versus runtime, camera versus evidence, implementation versus acceptance.

Applied AI · Agents · Voice · RAG

I build the orchestration, data contracts, APIs and safeguards that turn language models into working software.

Clinical decision support with governed knowledge, citations, deterministic application authority, and explicit human/production boundaries
Designed core LangGraph orchestration for a multi-tenant enterprise assistant: tenant-scoped retrieval, tools, state and failure handling
Multilingual speech and real-time interaction across API, worker, and browser-facing paths

Python · FastAPI · LangGraph/LangChain · Azure OpenAI · Qdrant/Milvus/pgvector · PostgreSQL · Playwright · Celery/RabbitMQ/Redis

Computer Vision · Video · Edge AI

Physical-world AI spanning camera/media input, perception, tracking, temporal state, evidence, target-device inference, and recovery.

Production multi-camera video intelligence across heterogeneous real-world sources and operational failure modes
Industrial and academic vision workflows for inspection, recognition, tracking, and workstation state
Edge optimization with TensorRT/CUDA/Jetson plus RGB-D and embedded perception work

PyTorch · TensorFlow · OpenCV · YOLO · ALPR · Tracking · TensorRT · CUDA · ONNX Runtime · Jetson · RTSP/HLS · FFmpeg

Quantitative ML · Credit Risk · Trading Systems

From credit-risk modeling to market research, I work on data availability, leakage, validation and production inference.

EPIC: Open Banking/CDR features with Spark/Iceberg, XGBoost/LightGBM validation and SageMaker inference
Curren: point-in-time market research, probabilistic forecasting and source-driven signal operations
Leakage-aware evaluation, reproducible data artifacts, selection controls, lifecycle/risk state, and reconciliation
Earlier equity, crypto, and FX research with ML ensembles, market data, cloud execution, and experiment tracking

Python · Rust · Spark · Iceberg · SageMaker · XGBoost · LightGBM · Arrow/Parquet · Polars · DuckDB · Scikit-learn · NautilusTrader

Research · Multimodal & Temporal ML

Peer-reviewed research across temporal video understanding, vision-language learning, medical time-series representation learning, and industrial computer vision.

Temporal action proposal generation and interaction modeling for long untrimmed video
Vision-language learning and coherent video paragraph captioning, including an AAAI 2023 Oral
Medical time-series representation learning and industrial visual inspection research

Transformers · Contrastive Learning · PyTorch · TensorFlow · Detectron2 · MATLAB · NumPy · SciPy · Scikit-learn · Weights & Biases

Flagship systems

Selected systems I've built and led.

Clinical AI, credit-risk ML, quant systems and production vision: the architecture choices, failures and release decisions behind the work. Confidential client data and proprietary implementation remain private.

Case study2026

AI Architect / Lead Builder · EPIC TECHNOLOGY

Clinical AI for a Traditional Medicine Hospital

EPIC TECHNOLOGY · Clinical AI & Decision Support

Built the clinical AI platform for prescription extraction, interaction and dose checks, plus source-grounded answers. FastAPI, hybrid Qdrant retrieval and deterministic clinical rules keep physician decisions separate from generated text.

Governed knowledgeDurable application stateExplicit clinical authority

Clinical AIAgentsRAGFastAPIPostgreSQLOn-prem

Open case study

Case studyNov 2025 — Present

Head of AI · Technical Lead · EPIC TECHNOLOGY

Credit Risk ML & Decision Infrastructure

EPIC TECHNOLOGY · Confidential Australian lender

Led an AWS-based credit-risk platform combining consent-qualified Open Banking data, point-in-time Spark/Iceberg features, XGBoost/LightGBM validation and SageMaker scoring. Corrected leakage and training/serving skew ahead of a staged production rollout.

Point-in-time featuresOffline/online parityRisk-governed deployment

Credit RiskAWSSparkIcebergSageMakerMLOps

Open case study

Case studyJun 2026 — Present

Independent AI/Quant Side Project · Solo Builder

Curren

Independent quantitative intelligence & trading-systems project

A solo-built quantitative research and systems platform connecting point-in-time evidence, reproducible evaluation, shared research/runtime semantics, guarded operational state, and a sanitized public verification surface — without publishing strategy internals.

Point-in-time evidenceResearch/runtime parityRead-only public surface

Quant ResearchRustPythonPIT DataTrading SystemsMCP

Open case study

Case study2025 — Present

AI / Computer Vision Systems Lead · EPIC TECHNOLOGY

Production Multi-Camera Video Intelligence

EPIC TECHNOLOGY · Computer Vision & Physical-World AI

The challenge beyond object detection: making heterogeneous cameras source-qualified, fresh, observable and recoverable, then connecting perception to trusted events and evidence on constrained hardware.

Source identity and freshnessTemporal detection and evidenceBounded recovery

Video AIComputer VisionYOLOALPRTrackingMedia pipelines

Open case study

Case study2024 — 2025

AI Engineer · FPT Software

Enterprise Conversational AI

FPT Software · Team-built agent systems

Designed and implemented core LangGraph orchestration for a multi-tenant enterprise agent platform, including tenant-aware retrieval, routing, tools, state, retries and response guards.

Typed agent stateTenant-aware retrievalTool and retry contracts

Enterprise AIAgentsLangGraphAzure OpenAIRAG

Open case study

Case study2024 — 2025

AI Engineer · FPT Software

Interview & Voice AI Systems

FPT Software · Conversation, speech & evaluation

Speech-enabled interview workflows integrating browser interaction, stateful scenarios, asynchronous AI services and evaluation, with explicit system-versus-personal ownership boundaries.

Voice interactionConversation/scenario stateAsync evaluation

Voice AIAzure SpeechNext.jsPythonInterview AI

Open case study

Selected system boundaries

Enough detail to prove depth. Not enough to leak the playbook.

These are intentionally abstracted architecture cuts: ownership, state, evidence, and failure boundaries are visible; sensitive implementation and operational details stay private.

Clinical AI

Governed clinical intelligence

EPIC TECHNOLOGY · AI Architect / Lead Builder

Clinician workflow AI reasoning Governed evidence Clinical authority

Shows the authority boundary, not the private corpus, prompts, infrastructure, or deployment topology.

Read the public case study

Physical-world AI

Video intelligence under real constraints

EPIC TECHNOLOGY · AI / Computer Vision Systems Lead

Media sources Perception Temporal state Evidence & recovery

The public view keeps vendor details, camera topology, credentials, and operational runbooks private.

Quant systems

Research discipline → guarded runtime

Independent side project · Solo Builder

PIT evidence Evaluation Runtime state Public verification

Strategy parameters, private data sources, execution controls, and trading authority are intentionally excluded.

Read the public case study

Research

Methods that survive beyond a paper

University of Arkansas · Applied Research

Research question Representation Evaluation Target hardware

Peer-reviewed work and deployment constraints are public; proprietary datasets and unsupported production claims are not.

System portfolio

More systems, from factory floors to enterprise agents.

Selected work across healthcare, education, enterprise agents, physical-world AI, quantitative systems, and research. Client/employer ownership stays explicit; sensitive implementation stays private.

01Nov 2025 — Present

EPIC TECHNOLOGY · Head of AI / Technical Lead

Credit Risk ML & Decision Infrastructure

Built the consent-aware Open Banking/Spark/Iceberg feature and SageMaker risk-scoring path for a confidential Australian lender. Temporal leakage, offline/online parity and lender approval were explicit rollout gates.

Production credit risk · AWS / MLOps

022026

EPIC TECHNOLOGY · AI Architect / Lead Builder

Hospital Clinical AI & Decision Support

Designed and deployed physician-facing clinical AI for prescription extraction, interaction/contraindication and dose checks, with source-grounded answers and deterministic medical rules.

Clinical AI · governed knowledge

032025 — Present

EPIC TECHNOLOGY · AI / Computer Vision Systems Lead

Production Multi-Camera Video Intelligence

Physical-world video AI spanning heterogeneous media sources, perception and tracking, temporal events, evidence, monitoring, and recovery under constrained hardware and vendor interfaces.

Production physical-world AI

042025 — Present

EPIC TECHNOLOGY · AI Architect / Lead Builder

AI-Powered Open edX Platform

Teacher-reviewable content generation and AI tutoring with retrieval, Vietnamese speech, real-time interaction, platform integration, browser verification, and operational delivery.

Education AI · human review

052025 — Present

EPIC TECHNOLOGY · AI / Computer Vision Engineering

Manufacturing Fastener Visual Inspection

Deployed edge YOLO inspection on one Japanese customer production line. Owned dataset preparation, fixture/ROI calibration and part-level decisions; handled false rejects after camera movement by recalibrating the vision path.

Production manufacturing · edge CV

062025 — Present

EPIC TECHNOLOGY · AI Systems Engineering

Internal AI Meeting Tool

Built and deployed an internally accepted meeting assistant with live speech, multi-turn conversation state and project-aware retrieval; addressed stale context and cross-project contamination.

Internal deployment · speech / RAG

072024 — 2025

FPT Software · AI Engineer

AI4U Enterprise Conversational Agent

Personally built core LangGraph orchestration for multi-tenant enterprise agents, including routing, tenant-aware retrieval, tool contracts and failure handling; wider platform delivery remained team-owned.

Enterprise agents · RAG

082024 — 2025

FPT Software · AI Engineer (team project)

Interview & Voice AI Systems

Speech-enabled interview and transcription workflows with browser voice interaction, conversational scenarios, asynchronous AI dialogue/evaluation and backend services.

Voice & interview AI · evaluation

092021 — 2024

University of Arkansas · Applied Research

Industrial & Poultry Vision Research

Visual inspection and workstation-state research connecting perception models to bounded physical-process evidence, alongside Jetson/TensorRT/CUDA edge optimization.

Industrial CV · edge inference

10Jun 2026 — Present

Independent side project · Solo Builder

Curren Quant Intelligence & Trading Systems

Point-in-time research evidence, reproducible evaluation, shared causal semantics, durable lifecycle/risk state, guarded execution boundaries, and sanitized public verification.

Research discipline · system parity

End-to-end ownership

Beyond the model call.

The recurring pattern is taking an uncertain research or product goal and making the ownership, system state, failure modes, evaluation, operations, and acceptance boundary explicit.

Architecture & authority

Turn ambiguous requirements into clear system boundaries, data ownership, interfaces, permissions, state, and acceptance criteria.

AI, tools & knowledge

Design model, retrieval, tool, memory, and human-review paths without letting probabilistic output silently become system truth.

Backend & product

Build the software around the model: APIs, durable state, async work, streaming, integrations, product surfaces, and operational controls.

Evaluation & reliability

Separate implemented from validated, make failure modes observable, and use evidence rather than fluent output as the acceptance boundary.

Cloud, edge & deployment

Carry systems through Docker/Linux, cloud or on-prem delivery, GPU/edge constraints, monitoring, recovery, and handoff.

Research to production

Preserve causality, provenance, reproducibility, and experimental assumptions while turning research into usable systems.

Career

Research depth. Product ownership. Production constraints.

My path runs from embedded computer vision and quant research to enterprise agents, credit-risk ML and production AI delivery.

Jun 2025 — Present

Head of Artificial Intelligence

EPIC TECHNOLOGY

Hands-on Head of AI leading architecture and implementation across production credit-risk ML, hospital clinical AI, video intelligence, manufacturing inspection and education AI.

Jun 2024 — Present

AI Tech Lead · Part-time / Consulting

A9 IOT

Project-based AI consulting, including FastAPI/LightGBM sensor forecasting with 16 metrics, 72 lag features and 36 recursive forecast steps.

Oct 2024 — Jun 2025

AI Engineer

FPT Software

Designed and implemented the core LangGraph AI orchestration for AI4U and contributed to multilingual speech/interview applications delivered by a wider FPT team.

May 2021 — Jan 2024

Graduate Researcher / PhD Student Researcher & Teaching Assistant

University of Arkansas

Graduate research in multimodal video understanding, vision-language learning, medical time-series learning, industrial computer vision, and edge inference; completed an MS in Computer Engineering.

Full chronology, teaching, skills and honors

Research

Peer-reviewed foundations.

Work across temporal video understanding, actor/object/environment interaction modeling, vision-language learning, medical time-series representation learning, and industrial computer vision. VLTinT was selected as an AAAI 2023 Oral.

Google Scholar

2024 CarcassFormer
Simultaneous localization, segmentation, and classification of poultry carcass defects Poultry Science 2023 AOE-Net
Entity-interaction modeling with adaptive attention for temporal action proposal generation International Journal of Computer Vision 2023 VLTinT
Visual-linguistic Transformer-in-Transformer for coherent video paragraph captioning · AAAI Oral AAAI 2023 sCL-ST
Supervised contrastive learning with semantic transformations for multi-lead ECG arrhythmia classification IEEE JBHI

Research rigor is only useful in production if the system preserves the contract that made the result valid.

That means explicit authority in clinical AI, causal evidence in quantitative systems, and source/failure awareness in physical-world AI.

Senior / lead roles

Need someone who can own the whole system?

I'm interested in applied AI leadership and senior engineering roles where model capability has to become reliable software: agent platforms, computer vision/video intelligence, multimodal AI, quantitative ML, distributed AI backends, and production delivery.

Start a conversation GitHub LinkedIn Full résumé