// Overview

Quick snapshot of who I am

const developer = {

Trinh The Minh

I am 

Building AI Products

From Concept to Production

Product-focused Full-Stack AI Engineer with proven expertise in the end-to-end lifecycle of AI applications. I build production-grade RAG, Knowledge Graph RAG (KG-RAG), and Multi-Agent systems.

};

class AIEngineer:

def build(self):

return "production_ready"

def deploy(self):

return "scalable_systems"

$ git log --oneline -5

a1b2c3d feat: deploy KG-RAG system

e4f5g6h fix: optimize vector search

i7j8k9l feat: add multi-agent pipeline

m0n1o2p perf: redis caching layer

q3r4s5t feat: auto review pipeline

// Projects

Production systems that drive real business impact

01

Parent Copilot - KG-RAG System

TeenUpSep 2025 - Dec 2025

Architected and launched a Knowledge Graph RAG using Neo4j and Vertex AI to power a 'Parent Copilot', driving a 30% DAU/MAU ratio through high-relevance AI interactions.

50%+ cost reduction by replacing legacy outsourced infrastructure
Deterministic Multi-Agent system with strict guardrails
Real-time user-facing chat with hallucination prevention
Neo4jVertex AIKnowledge GraphRAGMulti-Agent
02

Enterprise RAG Platform

Musashino Co., Ltd.April 2025 - Nov 2025

Deployed a self-hosted RAG system serving 50+ child companies, replacing Google's Vertex AI Agent Builder to eliminate unit limits and reduce operational costs by 40%.

Ultra-low latency semantic search with Redis caching
High concurrency handling without degradation
Automated MLOps workflows with Docker
RedisCeleryFastAPIVertex AIDocker
03

Auto Review Pipeline

PIXTA VietnamFeb 2025 - April 2025

Engineered the 'Auto Review' pipeline for Pixta Stock, processing 10,000+ daily images with 99% precision and 70% recall by fine-tuning CLIP and BLIP-v2 models.

10,000+ daily images processed
NER models for metadata extraction
Zero bottlenecks in content ingestion
CLIPBLIP-v2AWSDockerNER
04

Legal Documents Retrieval

SOICT Hackathon 2024Oct - Dec 2024

Developed a domain-specific retrieval engine using a hybrid Bi-Encoder and Cross-Encoder architecture, achieving MRR@10 of 0.7352.

Top 5 ranking among all participants
Hybrid retrieval architecture
Domain-specific optimization
Bi-EncoderCross-EncoderNLPRetrieval

// Technical Skills

Technologies I work with daily

skill_details.py

Hover over a skill to explore…

Built with by Trinh The Minh

© 2026 All rights reserved