Experience — Teetham Treelayapewat

Aug 2026 – Sep 2026Thailand · Onsite + remote

AltoTech

AI Engineer Intern

AI for building energy management.

Fine-tuned a 30B language model with LoRA for a 13-tool AI assistant running fully on-premises, without cloud inference.
Curated a 792-example English/Thai training corpus from real agent traces, including tool schemas and multi-turn tool calls.
Implemented hybrid RAG (vector + keyword) over 41 equipment manuals, improving retrieval from 8/12 to 10/12.
Expanded the agent evaluation suite to 289 cases testing tool selection, call arguments, retrieval and safety.
Profiled GPU memory and implemented a chunked cross-entropy loss to train on full 32k-token inputs on one 80 GB GPU.

Jun 2026 – Aug 2026Thailand · Remote

AI Researcher & Development Intern

Conversational AI company.

Delivered the store open/close detection module of a 5-module AI CCTV monitoring system for a retail client, combining YOLO, OCR and luminance analysis in a rule-based state machine.
Designed the proof-of-concept review interface for detection evidence and set up Docker Compose testing across 3 repos.
Reviewed 5 datasets and 7 papers to scope a Southeast Asian construction foundation model, then built door/window and room segmentation baselines on 400 floor plans.

Jun 2025 – May 2026Thailand · Onsite, then remote

AI Development Engineer Intern (extended to remote role)

3D-printed medical implants. Started onsite in Thailand, then kept working remotely through my first year at SUTD.

Developed a CT classification pipeline (DICOM → region classification → quality ranking) for the production CraNeXt system.
Reached 99.4% body-region accuracy (BiomedCLIP) and 96.1% craniectomy accuracy with a fine-tuned EVA-02, picked by benchmarking 8 models (5M–1,012M parameters) on accuracy, speed and size.
Achieved 1.12 mm error on cephalometric X-ray landmark detection with Thammasat University dentists, using multi-GPU training on the LANTA Supercomputer.