Ask about my work

An LLM answers on my behalf and can get things wrong. For anything important, contact me directly.

NatthakritBenjapatanamongkol

陳其骏

Student ID
m1561028
Degree
B.Eng., KMITL, 2024
Role
Freelance AI engineer
Location
Bangkok, Thailand

Education

  1. 2021 – 2024

    King Mongkut’s Institute of Technology Ladkrabang

    B.Eng. in Smart Materials Technology and Robotics and AI Engineering · International Program

    Courses
    Deep Learning · Computer Vision · Big Data · Cybersecurity
    Capstone
    Parking reservation system
  2. 2017 – 2020

    St. Paul’s School, Darjeeling, India

    Science–Mathematics

Experience

  1. 2025 – present

    King Mongkut’s Institute of Technology Ladkrabang

    Freelance AI engineer

    • AI-powered HR chatbot, lead developer. Built and tuned a full-stack RAG pipeline with Python, AWS Bedrock and a reranker model, then containerised it with Next.js, Python and Docker.
    • KMITL central lab chatbot, project lead. Led a two-person team building a RAG pipeline with a reranker over CSV files and website data, in Python on AWS Bedrock, with a Streamlit UI that can answer with images.
  2. Jul – Oct 2024

    G-able Company Ltd.

    AI engineer intern

    • ITSM chatbot. Built a multi-agent chatbot with LangChain for the IT service management department, with SQL query tools, API data retrieval and web search.
  3. Jun – Jul 2024

    Fukuoka Institute of Technology, Japan

    Internship program

    • Built a racing simulation in Unity for research and training.
    • Designed and programmed a line-tracking robot with embedded, real-time processing.
  • Python
  • LangChain
  • TensorFlow
  • PyTorch
  • OpenAI
  • YOLO
  • AWS Bedrock
  • Docker
  • Git
  • Next.js
  • HTML
  • CSS

Work

One research write-up, two live projects, and three more builds.

Research · In development

Focus: telling phone use from a phone on the desk

Can a webcam tell a phone you are using apart from a phone that is just lying on your desk?

  1. Session active
  2. Face present
  3. Phone detected
  4. Keyboard and mouse idle
  5. Phone movingunder test

All five hold

Held for the dwell time

Alert fires

Fig. 1 — All five must hold for the dwell time. The dashed one may be dropped.
Models
EfficientDet-Lite0 (float32) · BlazeFace · fully offline
Finding
The int8 model saw only noise on webcam frames. Float32 works, so Focus uses float32 only.
Open question
Is a phone moving in hand separable from the detector’s box jitter? If not, that check is dropped.

Project · LLM · RAG

NeoCare staff assistant

Answers questions from NeoCare staff and the service team.

Method
LLM + retrieval-augmented generation (RAG)
Intended use
Internal support for NeoCare staff.
Limits
Access is restricted to Thailand by Cloudflare; the demo will not load from other countries.

Project · Web app

Stash

Save links now, find them again when you need them.

Stack
Next.js · Drizzle ORM · Postgres · better-auth
Intended use
A searchable personal stash instead of a feed or a pile of tabs.
Limits
Requires a Google sign-in.

More projects

  • 2025

    Emotion classification

    Takes the text from a speech-to-text model and sorts it into three emotion categories with OpenAI GPT-4.

    Speech-to-text · GPT-4

  • 2024

    Parking reservation system

    Capstone. A Raspberry Pi runs the lot: ultrasonic sensors check spots in real time and OCR on the camera feed verifies each car. A Next.js interface takes reservations and checks licence plates against the backend.

    Raspberry Pi · ultrasonic · OCR · Next.js

  • 2023

    Driver drowsiness detection

    A Raspberry Pi and camera watch the driver and raise a real-time alert. A custom YOLO model trained on a labelled dataset detects visual signs of drowsiness with 97% accuracy.

    Raspberry Pi · YOLO