b
blueniev

Duong Thuyet

@blueniev

Fullstack Developer

Vietnã
Vietnamita, Inglês, Japonês
Algumas informações são exibidas no idioma inglês.
Sobre mim
I am a Fullstack Engineer specializing in Generative AI systems, web application with AI, and agentic architectures. I have experience building end-to-end AI pipelines and am committed to delivering impactful AI solutions.... Saiba mais

Habilidades

b
blueniev
Duong Thuyet
offline • 
Tempo médio de resposta: 12 horas

Conheça meus serviços

Desenvolvimento de chatbots de IA
I will build a rag ai chatbot trained on your documents or website content
Sites Personalizados
I will build a responsive react or nextjs frontend with typescript and tailwind

Portfólio

Experiência profissional

Amigo_Technologies JSC

AI Data Engineer

Amigo Technologies JSC

Apr 2026 - Present • 6 mos

• Built an internal policy assistant: a RAG agent that lets employees ask questions in natural language and get answers drawn from the company's own policy documents. • Turned the company's scanned internal PDFs into clean, structured Markdown with a purpose-built conversion tool, laying the data foundation for upcoming RAG and agent systems. • Built a dedicated search engine for the presale team that finds supporting evidence documents and generates content straight into the team's existing document templates.

Maxflow_Technology

Generative AI Engineer

Maxflow Technology

Aug 2025 - Dec 2025 • 4 mos

Job Description: - Research and develop a Tool Suggestion System based on Gen AI with a multi-stage retrieval pipeline - Research and implement an Agentic Translation System using a multi-agent architecture (LangGraph) - Optimize system performance and accuracy - Integrate and process complex documents (PDF, DOCX) Key Results: - Tool Suggestion System: Achieved 95.83% recall, 91.33% perfect recall, latency < 50ms. Modular system with 19 modules, integrating 5 retrieval methods and Learning-to-Rank. - Agentic Translation System: Document translation system with 6 agent nodes, supporting 50+ languages, integrating PDF in-place translation and DOCX style-preserving translation. - Optimization: Reduced init time from 30-150s to 1-2s, improved F1-score by +0.0011, reduced latency to 25-50ms. - PDF Processing: Fixed 3,504 empty TJ arrays to 0, ensuring PDF output has no blank pages, completed Phases 1-3 of the PDF stream module.