I am currently pursuing a dual degree in Information Technology Management at the American University of Phnom Penh and Computer Science at Fort Hays State University.

I am learning applied machine learning by building real products for Cambodian users — starting with Khmer speech recognition and agritech coordination. I am not a software engineer by identity; I focus on the ML and product side, and I work with a small team that owns the engineering.

• Khmer speech and language modeling — fine-tuning Whisper on Khmer audio, evaluating with CER, and building the data pipelines needed to reach usable quality.
• Applied ML for underserved sectors — forecasting, coordination, and early credit-scoring work for agriculture and informal commerce.
• Retrieval and agentic systems — embeddings, structured knowledge, and agent flows built on Claude and open models.
• AI product thinking — scoping features to what actually works, sizing markets honestly, and validating with real users before scale.
• Field research — farmer interviews, on-the-ground validation, and dataset design for low-resource Khmer contexts.

• Kaskor ASR — Khmer speech recognition model fine-tuned on ~86,000 audio samples using Whisper. Powers voice-first order intake within Chomkar. Still early — actively improving CER.
• Chomkar — Agritech coordination platform linking smallholder farmers in Kampong Cham to institutional buyers. Runner-up, Turing Hackathon Cycle 10. I led the ML roadmap and field validation; the team owns the engineering.
• The AI Layer (CL-00) — Shared AI/ML infrastructure and evaluation layer underpinning CHNAI LAB initiatives. In early development.
• Warden (SAT Digital) — Khmer-first Telegram moderation tool for Cambodian community operators. I contribute on the model design and evaluation side.

Python, PyTorch, Hugging Face, Whisper, Claude API, Cloudflare AI Gateway, Supabase, and PostgreSQL. I collaborate with teammates who own the frontend, backend, and infrastructure stacks.

I try to hold to strict claim discipline — describing outputs accurately, scoping features to what is actually shipped, and validating with real users before scale. I would rather ship something small and honest than something impressive-sounding and unverified. I am still early in the craft and expect to be for a long time.

• Portfolio: chamroeunhongleng.me
• Organization: @CHNAI-LAB
• From Kampong Cham, based in Phnom Penh