Nhu Vo  /  Projects

NLP systems, AI agents and the projects behind them

Products first: a medical speech translation prototype, two AI agent apps in development, and two live systems I managed. Below them, every research and training project I have worked on.

Public prototype · 1In development · 2Live system · 2All projects · 15
Research application

Medical speech translation

Built with the VinUni medical MT group.

Research application · Public prototype

Vietnamese–English Medical Speech Translation

My role
Project Coordinator
Problem
Doctors and patients who do not share a language struggle with spoken communication, especially when the conversation is full of medical terms.
Approach
The VinUni medical MT group chains Vietnamese speech recognition with Vietnamese–English medical machine translation in a real-time pipeline.
What it shows
  • A team product, not a solo one: Nguyen Giang Son built and hosts the MedComms demo on a server, running a speech-to-text (STT) model and a text-to-text translation model fine-tuned on the project's data.
  • Real-time pipeline: ASR → medical MT, served as a public web demo.
  • Built on the group's own data: MedEV (translation) and ViMedCSS (code-switched medical speech).
Status
A working team research prototype for clinical and educational settings. It is a research tool and has not been clinically validated.
Technology
Speech recognitionMedical machine translationVietnamese–English NLPModal (GPU)
Products

AI agent apps

Built with AI coding agents (Claude Code, Codex, Orca): I design, specify, test and review; the agents write most of the code.

No More Later · AI Priority Agent

In development
Role fitLLM / Agentic AI Engineer · AI Engineer
My role
Product owner & architect · AI-assisted build
Problem
To-do lists grow faster than anyone can work through them, and deciding what matters today takes more effort than the tasks themselves.
Approach
Turns tasks written in plain language into a daily plan: picks a Top 3, sorts work on an Eisenhower matrix, explains each priority, and replans when deadlines change. Every AI change waits for the user's approval.
What it shows
  • AI-assisted development with Claude Code and Codex, coordinated in Orca: I set architecture, specs and acceptance criteria, gated changes on tests and reviewed every merge; the agents wrote most of the code.
  • Multi-provider LLM gateway: OpenAI, Anthropic, Gemini and local Ollama, with cost-aware routing, fallbacks and a circuit breaker.
  • Prompt-injection test suite across every extraction workload; model output validated against JSON schemas.
  • Human-in-the-loop: assistant and replanning outputs are stored as proposals the user accepts or rejects.
Show 4 more
  • Per-user token budgets with atomic reservations; every inference logged with tokens and cost.
  • Privacy controls: per-user egress policy for cloud models, BYOK keys, SSRF-guarded outbound calls.
  • Semantic search with pgvector next to PostgreSQL full-text search; MCP server exposing task tools to other agents.
  • 160+ test files (Vitest), including database tests.
Status
A working monorepo with web, desktop and mobile clients, built with AI coding agents under human review. Public deployment is coming soon.
TypeScriptFastifyPostgreSQL · pgvectorpg-bossReactMCPOpenAI · Anthropic · Gemini · OllamaClaude Code · Codex · Orca

Togo Todo · AI Shared Planning Agent

In development
Role fitLLM / Agentic AI Engineer · AI Engineer
My role
Product owner & architect · AI-assisted build
Problem
Plans shared between several people, such as a trip or a joint project, end up scattered across chats, notes and maps.
Approach
A shared planning workspace with a chat agent: it reads a request, recommends activities, places and checklists, and only changes the shared plan after a member confirms.
What it shows
  • AI-assisted development with Claude Code and Codex, coordinated in Orca: I set architecture, specs and acceptance criteria, gated changes on tests and reviewed every merge; the agents wrote most of the code.
  • Multi-agent orchestration: an orchestrator routes each turn through language, ingest, RAG, places, policy, guardrail and mutation agents.
  • Hybrid RAG: pgvector similarity and full-text search merged with Reciprocal Rank Fusion.
  • Human-in-the-loop: every change becomes a pending action, validated again with Zod at confirm time.
Show 3 more
  • Agent observability: runs and spans stored in the database and traced in LangSmith; per-call LLM usage tracked.
  • Conversation memory per shared workspace; models via OpenRouter and local Ollama.
  • 89 test files: Vitest unit tests and Playwright end-to-end tests.
Status
Running locally with demo projects, built with AI coding agents under human review. Public deployment is coming soon.
TypeScriptNext.jsSupabase · PostgreSQL · pgvectorZodLangSmithOpenRouter · OllamaGoogle MapsClaude Code · Codex · Orca
Products

Institutional systems

Systems I managed from planning to launch with software partners. Both are live.

Green Transformation M&E System · Khanh Hoa

Live system
Role fitProject Manager
My role
Project Manager
Problem
Provincial departments need one place to monitor and evaluate progress on green transformation.
Approach
A website and mobile app for monitoring and evaluation, built with the software partner Bamboo Software (co-PI).
What it shows
  • 10 provincial departments involved.
  • PIs: Dr Thuy Bui, Nhu Vo and associates, at the Smart Green Transformation Center (GREEN-X), VinUniversity.
Status
Delivered Nov 2024 – Nov 2025 with 10 provincial departments of Khanh Hoa. Funded by VinUniversity.
Web applicationMobile appMonitoring & evaluation

VLE.HCMUE · E-learning System

Live system
Role fitProject Manager · E-learning
My role
HCMUE Team Leader
Problem
Teacher training under the Enhancing Teacher Education Program (ETEP) needed an online learning platform at HCMUE.
Approach
Built and rolled out the university's virtual learning environment with ACP Computer Training & Consultancy (Singapore).
What it shows
  • Funded by the Ministry of Education and Training and the World Bank.
  • HCMUE team: Dr Long Le, Ms Van Nguyen and Nhu Vo (team leader); ACP co-PIs: Alan Poh, Kelvin Lee.
Status
In use at HCMUE. Oct 2018 – Feb 2020; certificate awarded to the team leader for supporting the complete system.
E-learning platformOnline course design
Next

Planned

Ideas I plan to build next.

  • Coming soon
    Email Management AI Agent

    An LLM workflow that reads email, pulls out what matters, and drafts summaries and next actions for the user to review.

  • Concept
    LLM Translation Agent

    A workflow that moves from source text and instructions to an LLM draft, then to a terminology check and a review step, so errors are caught before the text is used.

  • Concept
    Speech Translation Agent

    Chain speech recognition, machine translation and LLM post-processing into a general-purpose speech translation assistant, extending the medical prototype beyond healthcare.

Project history

All projects

Research, product and training projects, newest first.

  • Dec 2025 – present
    Domain-specific Vietnamese Corpus Construction: EV Shift

    Co-investigator · Project Manager · VinUniversity

  • Jul 2025 – present
    Domain-specific Vietnamese Corpus Construction: Anti-Corruption

    Co-investigator · Project Manager · VinUniversity

  • Oct 2024 – present
  • Nov 2024 – Nov 2025
  • 2023 – present
    Twitter & ChatGPT project; Medical Chatbot project

    Research Assistant · VinUniversity

  • Aug 2022 – Apr 2023
    Natural Language Processing Engine (NLPE) for Natural-language Documents and Queries

    Co-investigator · Faculty of IT, HUFLIT

  • Apr 2022 – Apr 2023
    Corpus Construction of Business Speeches

    Co-investigator · VinUniversity

  • Apr 2021 – Apr 2022
    Building Vietnamese Learner English Corpus (ViLEC)

    Lead investigator · Faculty of Foreign Languages, NTTU

  • Feb 2020 – Apr 2021
    Building and Mining Vietnamese Corpus 1.0 (VnC 1.0)

    Lead investigator · Computational Linguistics Center, VNUHCM – University of Science

  • Oct 2018 – Feb 2020
    Building and Implementing the VLE.HCMUE E-learning System

    Team Leader (HCMUE) · Ministry of Education and Training & World Bank

  • Aug 2017 – Feb 2020
    Enhancing Teacher Education Program (ETEP)

    Trainer · Ministry of Education and Training & World Bank

  • Jun 2017 – Feb 2020
    Vietnam's National Foreign Language 2020 Project

    Trainer · Ministry of Education and Training

  • Oct 2018 – Aug 2019
    Short-term Courses Construction

    Team Leader · IT Center, HCMC University of Education

  • Jan 2018 – Dec 2019
    Building Pedagogical Scenarios for Short-Term Online Courses (CS.2018.19.08)

    Member · HCMC University of Education

  • Jan 2017 – Dec 2019
    Building MOOCs from a Positive Teaching Perspective (CS.2017.19.03TK)

    Member · HCMC University of Education