AI engineer working across speech systems, LLM applications, and the backend infrastructure that makes them usable in production.

I have spent years building around ASR, TTS, callbot/chatbot systems, LangChain and LangGraph agents, RAG, realtime audio, and practical automation. I care about systems that are fast enough, observable enough, and simple enough to keep running after the demo.

Ho Chi Minh City, Vietnam
GitHub: https://github.com/thanhdat77

• Speech AI: ASR, TTS, VAD pipelines, realtime audio over WebSocket, callbot flows
• LLM applications: LangChain, LangGraph, tool calling, RAG, stateful agents
• Backend systems: FastAPI, async Python, Kafka, Redis, MongoDB, Qdrant, MinIO
• Production workflow: Docker, uv, pytest, ruff, observability, API contracts
• Developer environment: Neovim, tmux, dotfiles, Pi agent workflows, custom keyboards

Event-driven callbot services that connect ASR, NLU/LLM logic, TTS, session state, and business actions.

• WebSocket gateway for realtime audio sessions
• Kafka-based ASR/TTS/event routing
• Redis-backed hot state, cache, and coordination
• Vietnamese text normalization for speech output
• Graph-style call flows with interrupt, fallback, and action handling

Tech: Python, FastAPI, WebSocket, Kafka, Redis, Docker

LangGraph-based chatbot architecture for multi-turn customer workflows, retrieval, state, and backend-facing actions.

• LangGraph agent runtime with conversation state and routing
• RAG over Qdrant with document ingestion and reranking paths
• MongoDB/Beanie for durable data, Redis for hot state and checkpointing
• SSE/streaming API surface for frontend and backend integration
• Provider-wrapped LLMs and embeddings

Tech: Python 3.12, FastAPI, LangGraph, LangChain, Qdrant, MongoDB, Redis

ASR API service for uploaded files, local media, MinIO objects, and raw bytes.

• VAD segmentation before ASR
• Remote ASR worker pool over WebSocket
• Sync and async callback modes
• Redis caching and MinIO integration
• Public and legacy API contracts

Tech: FastAPI, WebSocket, Redis, MinIO, Docker, pytest

TTS API and test harnesses for comparing providers, saving generated WAV files, and measuring runtime behavior.

• Queue-based TTS API with callback and polling support
• Provider-agnostic TTS testkit
• Local and remote provider adapters
• Experiments with F5-TTS, VITS, Kokoro, StyleTTS2, OmniVoice/Triton
• GPU-aware benchmarking and Dockerized model runs

Tech: Python, FastAPI, Redis, ONNX Runtime, Triton, CUDA-aware workflows

Languages

AI and LLM

Backend and Data

Tools

• I tune my workflow around Neovim, terminal tools, and dotfiles.
• I like experimenting with Pi agent style automation for day-to-day coding.
• I enjoy custom keyboards, firmware/configs, and efficient layouts.