iGOT.ai: AI agents and workflow platform

The graph engine behind a no-code AI agent builder, then leading the team that built five AI and Web3 products on top of it.

Company
iGOT.aiNo-code platform for building LLM-powered virtual assistants and workflow agents.
Role
Frontend Engineer, then Engineering Manager
Period
Mar 2024 – Mar 2025
Location
Vietnam (remote)

Original reconstruction for this portfolio.

Overview

iGOT.ai is a no-code platform for building LLM-powered assistants and workflow agents. I joined as a frontend engineer and architected the graph engine behind iGOT.ai Studio’s agent builder.

Six months later I became engineering manager, leading 7 engineers across 5 AI and Web3 products, including a multi-agent RAG platform for a government client.

Context and problem

In a no-code agent builder, the canvas is the program. If what people draw and what the backend runs disagree, the product is wrong in a way users can’t debug.

As a manager, the problem shifted from one engine to five products that needed consistent builds, reviews, and delivery.

My role and ownership

What I built

  • Architected the graph engine behind iGOT.ai Studio’s no-code AI agent builder in React Flow, syncing canvas edges with backend execution order.
  • Added auto-layout and optimistic node creation; the engine powers 23 task-node types.
  • Developed the iGOT.ai mobile app with voice-based AI interaction in React Native, and published the @igot.ai/react-native npm package.
  • Implemented a Siri-like voice assistant for EMDDI, a ride-hailing app, and a Flutter chat-bubble boilerplate to embed iGOT.ai agents in iHanoi.

What I led

  • Managed and mentored 7 engineers across 5 AI and Web3 products; standardized containerized builds, CI/CD, and pre-commit quality gates.
  • Architected and shipped a full-stack AI audit platform for a Vietnamese government client: a 6-stage multi-agent RAG pipeline, role-based access, and versioned AI results with a full audit trail.
  • Architected and led Workbook, a NotebookLM-style research app with multi-agent workflows and Google Docs/Sheets/Drive sources, built with Python, FastAPI, and PostgreSQL.
  • Led delivery of Datalog (formerly Catalog): owned the project plan and ran architecture walkthroughs and code reviews.
  • Led development of OpenWare, a Solana DeFi platform with an AI-powered token launcher and a real-time multi-wallet trading interface.

Challenges

  • The canvas is the program

    Edges drawn on the canvas had to match the order the backend actually executes tasks in.

  • Graphs that stay readable

    As people add nodes, layouts tangle. New nodes also need to appear instantly, before the server confirms them.

  • Five products, one team

    Different stacks and deadlines needed shared standards for builds, CI/CD, and quality.

Design and architecture decisions

  • Edges define execution order

    The graph engine keeps canvas edges and backend execution order in sync, so what users draw is what runs.

  • Auto-layout and optimistic nodes

    Layout runs automatically and new nodes appear immediately, then reconcile with the backend.

  • Standardize the delivery path

    Containerized builds, CI/CD, and pre-commit quality gates across the team’s products.

  • Audit everything the AI decides

    The government platform keeps versioned AI results with a full audit trail, behind role-based access.

Selected implementation details

Loading the interactive reconstruction…

  • Graph engine

    React Flow canvas with edge-to-execution-order sync, auto-layout, and optimistic node creation for 23 task-node types.

  • AI audit platform

    6-stage multi-agent RAG pipeline, role-based access, versioned results, and an audit trail.

  • Mobile and embeds

    React Native voice app, the @igot.ai/react-native package, and a Flutter chat bubble for iHanoi.

Outcomes

23task-node types on one graph engine
iGOT.ai Studio’s no-code agent builder.
7engineers managed and mentored
Across 5 AI and Web3 products, as engineering manager.
6stage multi-agent RAG pipeline
Shipped for a Vietnamese government client, with a full audit trail.

Figures come from my resume or from measuring the shipped product. Where a number describes the product or the team rather than my work, the context says so.

Technologies

  • React
  • React Flow
  • TypeScript
  • Python
  • FastAPI
  • PostgreSQL
  • React Native
  • Flutter
  • Docker
  • CI/CD