Post Job Free
Sign in

AI/ML Engineer with Generative AI Focus

Location:
Long Hoa, Vinh Long, Vietnam
Posted:
October 30, 2026

Contact this candidate

Resume:

PHAN MINH NHAT TRUONG

AI/ML ENGINEER GENERATIVE AI APPLIED ML ENGINEER

Ho Chi Minh City, Vietnam +84-918-***-*** *********@*****.*** CAREER OBJECTIVE

Artificial Intelligence undergraduate with hands-on experience building applications across machine learning, computer vision, NLP, and LLM-powered generative AI. Seeking an AI/ML Engineer role to contribute to practical model development, system integration, and reliable deployment while strengthening expertise through real-world engineering challenges. EDUCATION

FPT University 2023 – Present

Bachelor’s Program in Artificial Intelligence Ho Chi Minh City, Vietnam HONORS & AWARDS

• 1st Place – JetRacer AI Racer Challenge 2026 (Aita Lab). Built an autonomous driving system for NVIDIA Jetson Nano combining real-time perception and control. [Project] TECHNICAL SKILLS

Programming: Python, R, SQL, JavaScript/TypeScript Machine Learning & GenAI: PyTorch, Diffusers, scikit-learn, LoRA/PEFT NLP & LLMs: LLM API Integration, LangGraph, RAG, Sentence Transformers, Text Embeddings, Prompt Engineering Computer Vision & Edge AI: OpenCV, YOLOv8, ONNX Runtime, TensorRT, ROS Database & Storage: FAISS indexing, PostgreSQL, DynamoDB, Amazon S3 Vectors Cloud & Backend: AWS, GCP, Modal, FastAPI

Development Tools: Git, Docker, Linux

EXPERIENCE

AWS Study Group – First Cloud AI Journey Workforce Bootcamp Jan 2026 – Apr 2026 Cloud & AI Intern (Bootcamp) Ho Chi Minh City, Vietnam

• Built Voice Summarizer, a React/FastAPI application for transcribing recordings, summarizing discussions, and answering questions with LLM-based RAG over transcript content. [GitHub]

• Designed an asynchronous processing workflow: direct browser uploads to S3 via presigned URLs, Lambda-triggered transcription with Amazon Transcribe, and Celery/Redis workers for summarization and indexing; tracked recording status in DynamoDB.

• Segmented transcripts by semantic topic shifts using Sentence Transformers and indexed chunks in S3 Vectors. Routed questions to relevant passages, topic summaries, or recording-wide summaries, with metadata filters restricting retrieval to the selected recording.

PROJECTS

StoryMapDiffusion – Generative Geospatial AI Platform [GitHub] [Live Demo] Solo Project Qwen-Image-Edit-2511, PyTorch, LoRA, Modal, FastAPI Transforms 3D city captures into stylized, zoomable maps with an AI guide for landmark exploration.

• Built a full-stack geospatial GenAI workflow from GeoJSON-defined 3D city capture and render-quality checks to paired- dataset preparation, remote GPU inference, and interactive Deep Zoom maps.

• Fine-tuned Qwen-Image-Edit-2511 using a custom BF16 LoRA pipeline with coordinate-aligned render/style pairs, mask- augmented editing inputs, and resumable training checkpoints.

• Optimized A100 80GB inference with 4-step Lightning, fused LoRA, and PyTorch compilation; measured 25.5 s median GPU processing and 34.9 s median client streaming time per 8-image request.

• Built an asynchronous client–GPU server pipeline with Base64-encoded image payloads and NDJSON result streaming; prefetched next-batch inputs during inference, encoded results off the server event loop, and overlapped result reception with local image commits through a bounded queue.

• Generated 3,531 map tiles (3.70 gigapixels) in 16.5 hours for about US$39.74 in GPU costs, with resumable batch processing and multiresolution DZI export.

• Built an LLM-powered landmark guide with RAG over Wikipedia/Wikidata to explain landmarks and recommend related places. Combined geographic coordinates, camera and tile metadata, terrain elevation, and building heights to project recommendation pins onto the 2D map with height and occlusion adjustments. JetRacer AI – Autonomous Driving & Edge AI [GitHub] 1st Place, JetRacer AI Racer Challenge 2026 ROS, TensorRT, Jetson Nano A vision-based driving system for a physical JetRacer robot, supporting track racing and navigation through traffic signals and obstacles.

• Deployed ONNX models with TensorRT FP16 and cached engines on Jetson Nano; achieved approximately 68 FPS average in track mode and 44 FPS average with three models running in urban mode.

• Combined YOLOv8 traffic-sign/light detection and MobileNet obstruction classification with route-conditioned trajectory prediction; fused visual features with learned turn-command embeddings to predict waypoints for intersection navigation.

• Connected ResNet-18 lane predictions to Stanley steering and predicted waypoints to Pure Pursuit control. Applied Kalman filtering, steering-dependent throttle, and an obstacle-recovery state machine with smoothed detection scores and separate entry/exit thresholds to suppress unstable decisions. Comic Studio – Agentic AI Comic Generation & Editing [GitHub] Solo Project LangGraph, Stable Diffusion 1.5, FastAPI, React Turns story ideas into editable comic pages with generated artwork, panel layouts, and English/Vietnamese dialogue.

• Orchestrated LLM story planning, dialogue, storyboarding, and structured validation with LangGraph; used a vision-language model to compare rendered pages against the story and route corrections to the relevant planning or rendering stage.

• Supported detailed prompts beyond CLIP’s 77-token context window through Compel’s chunked, weighted text encoding; matched positive and negative embedding lengths for Stable Diffusion inference with FP16 and style LoRAs.

• Automated speech-bubble placement using foreground segmentation, prioritized free-space regions, and rectangle-overlap checks; sized bubbles from font measurements and used grid-search fallback to reduce character occlusion and bubble collisions.

• Built a React/FastAPI editor with selective panel regeneration and progress updates streamed through server-sent events

(SSE); reused unaffected artwork and applied dialogue-only edits without diffusion inference. CERTIFICATIONS

• Probabilistic Graphical Models Specialization – Stanford Online (2026)

• Deep Learning Specialization – DeepLearning.AI (2025)

• Retrieval Augmented Generation – DeepLearning.AI (2025)

• Data Science Specialization – Johns Hopkins University (2025)



Contact this candidate