Vector Nguyen

AI engineer building multi-agent systems, LLM products and agentic voice.

Hi, I'm Vector. At MoMo I designed and built Tro Ly An Choi, a zero-to-one AI chat assistant with 1 orchestrator and 6 domain agents. It runs across 10+ MoMo mini-app surfaces and reaches over 100K monthly engaged users. I built 70+ of its agent tools, covering cinema, airline, bus and train ticketing and food ordering, all with deep payment integration into MoMo's checkout.

I studied Mechatronics Engineering at HCMUTE, then moved into AI: first at FPT Software building computer vision and NLP systems, then at FPT's AI Center on multimodal RAG, multi-agent architectures and multilingual AI for enterprise clients. More about me

Building

Assistant Core

Jan 2024 – present

Deploy one AI assistant everywhere — web, mobile widgets, IoT devices, wearables and embedded surfaces — from a single configuration, without duplicating logic per channel.

  • Multi-provider: 7+ AI providers (OpenAI, Anthropic, Google, xAI, DeepSeek, AWS Bedrock, ElevenLabs, Soniox), switchable without reconfiguration.
  • Voice pipeline: sub-1s ASR → LLM → TTS streaming with VAD auto mode, barge-in interruption and wake-word customization over WebSocket and MQTT+UDP.
  • Knowledge & tools: agentic RAG over documents, websites and databases; MCP server integration; custom API tool calling.
  • White-label multi-tenant: RBAC, custom domain routing, and an admin dashboard for assistant config, users, device pairing and voice quotas.
  • PostgreSQL/pgvector
  • Redis
  • S3
  • Langfuse
  • Grafana
  • TLS 1.3
  • JWT/OAuth 2.0

assistantcore.com

Updates

All updates

  1. Won Champions of Hacking Week at Tech Day MoMo 2025.

  2. Joined MoMo as an AI Engineer and built Tro Ly An Choi from zero — a multi-agent chat assistant (1 orchestrator + 6 domain agents) integrated across 10+ MoMo mini-app surfaces, reaching over 100K monthly engaged users.

  3. Moved to Ho Chi Minh City to join FPT Software AI Center as an AI Engineer, working on LLMs, multi-agent architectures and multimodal RAG.

  4. Received the Most Valuable Player 2022 award at FPT Software.

  5. Won Second Prize at the Quy Nhon AI Hackathon 2022.

Side projects

More on GitHub

Search Engine System

Dec 2023

Scalable image and full-text search over 100,000 Shopee products, built as microservices.

  • Image search with async programming, gRPC, a vector database, TensorRT and dynamic batching — 200 RPS at 100 ms p95 latency.
  • Full-text search and autocomplete on Elasticsearch with custom scoring scripts, improving relevancy by 35%.
  • CI/CD with Docker, GitHub Actions, Ansible and CloudFormation; deployed to AWS EKS with Helm charts.
  • gRPC
  • TensorRT
  • Elasticsearch
  • Vector DB
  • AWS EKS

Source

Face Recognition System

Sep 2023

Real-time face recognition across six Dockerized microservices with GPU inference.

  • SCRFD detection, ByteTrack multi-object tracking and ArcFace embeddings, all served by NVIDIA Triton Inference Server.
  • Few-shot enrollment with ArcFace embeddings and Qdrant similarity search; Redis Streams carry frames and boxes between services.
  • Served via FastAPI with an MJPEG live video feed.
  • Triton
  • ArcFace
  • ByteTrack
  • Qdrant
  • Redis Streams

Source

JobFit: AI-powered resume speed screening

Resume Ranking

Apr 2023

Ranks resumes using the language understanding of large language models.

  • Prompt engineering extracts key insights from each resume and produces accurate rankings.
  • LLMs
  • Prompt engineering

Source Demo video