🚀 AI Engineer | Computer Vision | LLM Systems

• A multi-agent AI platform for Vietnamese health insurance that supports policy Q&A, product comparison, and claim advisory with evidence-first RAG. It builds an end-to-end PDF-to-RAG pipeline that converts insurance PDFs into Markdown, transforms complex benefit and premium tables into narrative text before embedding, and indexes structured/unstructured content into SQLite, Qdrant, and Knowledge Graph layers; normalizes a traceable database from 6 insurers, 83 ingested PDFs, 644 source tables, 424 plans, 3,766 benefit values, 7,004 hospitals, and 551 claim payout records; includes a read-only SQLite MCP server with 21 insurance tools, a LangChain DatabaseAgent with Langfuse tracing, an AI-assisted graph schema discovery pipeline (42 node types, 68 relationship types, 127 node properties, 27 relationship properties), and a document-grounded chunking benchmark where the best strategy reached 62.00% Primary Hit@5.
• Tech Stack & Skills: FastAPI, Knowledge Graph, LangChain, Langfuse, MCP/FastMCP, Multi-Agent Systems, Neo4j, NetworkX, Python, PyTorch/CUDA, Qdrant, RAG, SQLite
• Source: InsureVN

• An all-in-one AI Meeting Assistant tailored for the Vietnamese language and workplace. It automates the entire meeting lifecycle—from real-time transcription and speaker diarization to AI-powered summarization and action item extraction. It features a RAG-based knowledge base that transforms transcripts and uploaded documents (PDF/Word) into a searchable intelligence hub, enabling users to query meeting history and cross-reference data with high precision.
• Tech Stack & Skills: Docker, FastAPI, Gemini, Next.js, Node.js, PostgreSQL/pgvector, Python, RAG, React, Supabase, Tailwind CSS
• Source: IQMEET-Meeting-Intelligence-Assistant

• A production-grade Agentic Temporal GraphRAG system designed for complex meeting intelligence. Built for AI engineers and businesses needing to extract accurate relationships, timelines, and insights from meeting transcripts. Achieves <1s latency with hybrid search and Reciprocal Rank Fusion (RRF) while supporting multi-provider LLMs.
• Tech Stack & Skills: Docker, FastAPI, LLM, LangChain, Neo4j, Python, Qdrant
• Source: IMA-Assistance-Agent-RAG

• Extract deep, actionable insights from thousands of meeting transcripts with a local-first Agentic Temporal GraphRAG pipeline. Built for enterprise teams, analysts, and project managers who need to track decisions, commitments, and topics over time. Combines LangChain, Neo4j knowledge graphs, and Qdrant vector search to map the exact "who, what, and when" of your organization's meetings.
• Tech Stack & Skills: Docker, FastAPI, LangChain, Neo4j, Python, Qdrant, RAG
• Source: iqmeet-agentic-rag

• Automate your Google Workspace workflows safely with a smart AI agent featuring built-in Human-in-the-Loop (HITL) approval. Built for teams and professionals who want to delegate tedious tasks like email triaging, document summarization, and calendar management without losing control over critical actions. Safely execute 100+ Workspace actions with 1-click approvals.
• Tech Stack & Skills: FastAPI, Python
• Source: gworkspace-ai-agent

• Extract, embed, and query complex documents instantly with our robust Retrieval-Augmented Generation (RAG) engine. Built for data scientists, researchers, and knowledge workers who need accurate answers from massive PDF troves, it effortlessly processes over 100 pages per minute with high-fidelity semantic search.
• Tech Stack & Skills: Accelerate, Hugging Face, LLM, LanceDB, PyMuPDF, PyTorch, Python, RAG, SentenceTransformers, Transformers
• Source: Askly

• Provide empathetic, 24/7 mental health support using an intelligent Retrieval-Augmented Generation (RAG) chatbot. Designed for wellness platforms, counseling centers, and health tech developers, PawsitiveMind leverages LanceDB and the lightning-fast Groq API to deliver sub-second, contextually accurate responses grounded in a vast library of therapeutic dialogues.
• Tech Stack & Skills: Flask, Groq, LLM, LanceDB, PyTorch, Python, RAG, Streamlit
• Source: Chatbot_mentalhealth_cat-dog

• A high-performance RAG-powered chatbot delivering accurate pet health advice and care guidelines. Built for pet owners and veterinary clinics needing instant, reliable answers backed by LanceDB vector search. Handles thousands of queries securely with high precision and sub-second retrieval times.
• Tech Stack & Skills: Flask, Groq, LLM, LanceDB, PyTorch, Python, RAG, SentenceTransformers
• Source: pet-health-assistant

• Deliver personalized learning experiences at scale with a robust, AI-driven educational backend. Engineered for EdTech startups, universities, and online academies, Edulight seamlessly handles thousands of concurrent students while processing heavy AI generation tasks in the background, boasting a 99.9% uptime architecture.
• Tech Stack & Skills: Celery, Docker, FastAPI, JavaScript, LLM, PostgreSQL, Python, Redis, Supabase
• Source: edulight

• A powerful, fine-tuned adaptation of the Qwen3-ASR model specifically optimized for Vietnamese speech recognition and alignment. Built for AI researchers and developers needing high-accuracy transcription, dialect support, and forced alignment for complex Vietnamese audio. Achieves exceptional Word Error Rates (WER) on regional benchmarks and scales easily with vLLM backends.
• Tech Stack & Skills: Docker, Gradio, Hugging Face, Python, Transformers, vLLM
• Source: Qwen3-ASR-VI

• Automate the tedious process of video dubbing with a fully integrated, open-source pipeline. Built for content creators, educators, and media companies who need to localize video content rapidly without losing quality. Fully automate video dubbing into Vietnamese in 5 seamless steps with zero audio-visual desync.
• Tech Stack & Skills: Docker, FFmpeg, PyTorch, Python
• Source: VoiceWeaver

• End-to-end automated highlight clip generation for full soccer matches. Built for sports analysts, broadcasters, and football fans who want to extract ranked, key moments (goals, shots, passes) in minutes instead of manually scrubbing through hours of footage.
• Tech Stack & Skills: Accelerate, Docker, FFmpeg, FastAPI, Gradio, OpenCV, PyTorch, Python
• Source: soccer-highlight

• An end-to-end toolkit for training, evaluating, and deploying state-of-the-art brand logo segmentation models. Built for marketing analysts, broadcast monitors, and computer vision researchers who need pixel-perfect logo identification. Detect and segment brand logos in real-time at 60 FPS with 94.4% mAP using the power of YOLO11.
• Tech Stack & Skills: Computer Vision, Docker, Gradio, Hugging Face, OpenCV, PyTorch, Python, YOLO, YOLO11
• Source: yolo11-seg-logo

• Automate and refine image segmentation masks instantly with YOLO and an intuitive web interface. Tailored for computer vision researchers and data annotators, AutoMask Refinery speeds up dataset preparation by up to 10x by combining zero-shot AI predictions with responsive human-in-the-loop correction tools.
• Tech Stack & Skills: Computer Vision, Flask, OpenCV, PyTorch, Python, YOLO
• Source: AutoMask-Refinery

• This project uses YOLOv26 Segmentation to automatically detect and mask betting logos (e.g., 1xbet, melbet, etc.) in videos.
• Tech Stack & Skills: Docker, FastAPI, OpenCV, Python, YOLO
• Source: Smart-Logo-Masker

• A comprehensive toolkit designed to streamline the creation of high-quality Automatic Speech Recognition (ASR) datasets directly from YouTube. Built for machine learning engineers and linguists who need massive, clean audio data. Build 1,000+ hour ASR datasets automatically with integrated Voice Activity Detection (VAD) and real-time dashboard monitoring.
• Tech Stack & Skills: AWS, Hugging Face, PyTorch, Python, Streamlit
• Source: yt-asr-kit

• Break down language barriers with high-fidelity, culturally aware translations powered by the cutting-edge LLaMA 3.1 models. Built for content creators, localization teams, and global businesses who need nuanced translations beyond what standard tools provide. Translate content across 50+ languages with culturally aware nuance using the massive 405B parameter model.
• Tech Stack & Skills: Python, Streamlit
• Source: Translations-with-LLaMA

• Generate vivid, real-time, and thrilling live commentary for horse racing videos using Google Gemini 2.5 Flash. Built for sports broadcasters, content creators, and developers who need automated, contextual, and hyper-realistic audio/text descriptions of fast-paced sporting events. Process unlimited video chunks with continuous memory of the race state.
• Tech Stack & Skills: Python, ML/AI
• Source: horce_racing

• Optimize your CV instantly using advanced AI to perfectly match job descriptions. Built for job seekers and career coaches who need to increase interview callback rates by up to 300% with automated, intelligent resume tailoring.
• Tech Stack & Skills: Docker, Groq, LLM, Python, Streamlit
• Source: ResumeEnhancer

• Master the TOEIC exam with a personalized, AI-driven study companion that adapts to your learning pace and targets your weak points. Built for language learners and educators who demand a smart, interactive, and modern approach to test preparation. Boost your TOEIC score by 150+ points in just 4 weeks of consistent study.
• Tech Stack & Skills: GCP, JavaScript, React, TypeScript, Vite
• Source: toeic_master

• Programming & Frameworks: Python, FastAPI, PyTorch, Hugging Face Transformers.
• Generative AI & LLM Engineering: LLMs (OpenAI, Gemini, Ollama, Llama.cpp), Fine-tuning (LoRA, QLoRA), Quantization (GGUF, EXL2, AWQ), LLM Evaluation (RAGAS, G-Eval, DeepEval), Prompt Engineering, RAG Pipelines, Agentic AI (Multi-agent workflows, Tool calling).
• AI Frameworks & Tools: LangChain, LlamaIndex, AutoGen, Pydantic, Gradio, Ultralytics, LiteLLM, Vector Databases (Qdrant), Hugging Face Transformers.
• Machine Learning & CV: PyTorch, TensorFlow, Scikit-learn, CNN, RNN, Transformers, Computer Vision (Object Detection, OCR, Video Analysis), NLP (Semantic Search, Text Classification), Recommendation Systems, Anomaly Detection.
• MLOps: Docker, CI/CD (GitHub Actions), Model Serving & Monitoring, Google Analytics.
• QA Automation & Testing: Playwright, Selenium, Cypress, AI-assisted Testing (Test generation, Self-healing scripts), Unit/Integration/Regression Testing.
• Language Skills: TOEIC: 650+

AI is software, not magic.