| --- |
| title: icdevelop |
| emoji: 🧬 |
| colorFrom: indigo |
| colorTo: blue |
| sdk: static |
| pinned: false |
| --- |
| |
| <style> |
| .ic-wrap { max-width: 1100px; } |
| .ic-lead { font-size: 1.08rem; line-height: 1.65; } |
| .ic-grid { display: grid; gap: 16px; margin: 20px 0; } |
| .ic-g2 { grid-template-columns: repeat(auto-fit, minmax(280px, 1fr)); } |
| .ic-g4 { grid-template-columns: repeat(auto-fit, minmax(160px, 1fr)); } |
| .ic-card { |
| border: 1px solid rgba(128,128,128,.28); |
| border-radius: 12px; |
| padding: 18px 20px; |
| background: rgba(128,128,128,.05); |
| } |
| .ic-card h3 { margin: 0 0 6px 0; font-size: 1.05rem; font-weight: 700; } |
| .ic-card p { margin: 0 0 10px 0; font-size: .93rem; line-height: 1.55; opacity: .88; } |
| .ic-kicker { font-size: .72rem; letter-spacing: .12em; text-transform: uppercase; |
| color: #4F86F7; font-weight: 700; margin: 0 0 8px 0; } |
| .ic-stat { border-left: 3px solid #4F86F7; padding: 4px 0 4px 14px; } |
| .ic-stat .n { display: block; font-size: 1.55rem; font-weight: 700; line-height: 1.2; } |
| .ic-stat .l { display: block; font-size: .78rem; opacity: .72; } |
| .ic-links a { margin-right: 14px; font-size: .9rem; white-space: nowrap; } |
| .ic-note { border-left: 3px solid #C2410C; padding: 2px 0 2px 14px; font-size: .92rem; opacity: .9; } |
| </style> |
| |
| <div class="ic-wrap"> |
|
|
| <img src="banner.png" alt="icdevelop — Domain-Specific LLMs for Thai Professionals" /> |
|
|
| <p class="ic-lead"> |
| We build <b>open-weight Thai language models for regulated professional work</b> — models that a |
| pharmacist, a nurse or a field engineer can actually put in front of a task, not general chatbots |
| with a Thai accent. Every release ships with the benchmark it was gated on, the limits we measured, |
| and an acknowledgement form the user has to read first. |
| </p> |
|
|
| <div class="ic-grid ic-g4"> |
| <div class="ic-stat"><span class="n">6</span><span class="l">open-weight models released</span></div> |
| <div class="ic-stat"><span class="n">2</span><span class="l">curated Thai knowledge bases</span></div> |
| <div class="ic-stat"><span class="n">90.0%</span><span class="l">knowledge + RAG · TMA-1 27B FP8</span></div> |
| <div class="ic-stat"><span class="n">98.3%</span><span class="l">safety / referral · TMA-1 27B</span></div> |
| </div> |
|
|
| <p class="ic-links"> |
| <a href="https://huggingface.co/collections/icdevelop/thai-medical-assistant-6a695b2455f4aadc5f68fa7e"><b>Thai Medical Assistant →</b></a> |
| <a href="https://huggingface.co/collections/icdevelop/thai-pharmacy-assistant-6a5c815879e683e9ae9c697a"><b>Thai Pharmacy Assistant →</b></a> |
| <a href="https://www.i-c-develop.com">i-c-develop.com</a> |
| <a href="https://www.facebook.com/ICDevelop">Facebook</a> |
| </p> |
|
|
| <hr> |
|
|
| <p class="ic-kicker">Model lines</p> |
|
|
| <div class="ic-grid ic-g2"> |
| <div class="ic-card"> |
| <h3>TMA-1 · Thai Medical Assistant</h3> |
| <p>Thai consumer-health assistant: takes a short history, rules out options that are unsafe for |
| that person <i>with reasons</i>, refers on red flags, and refuses to diagnose. Multimodal — |
| medicine boxes, labels and home-device readouts, not radiology. Built to run with RAG over a |
| curated Thai medical knowledge base.</p> |
| <p><a href="https://huggingface.co/icdevelop/tma1-medgemma-27b-fp8">27B FP8</a> · |
| <a href="https://huggingface.co/icdevelop/tma1-medgemma-27b">27B</a> · |
| <a href="https://huggingface.co/icdevelop/tma1-medgemma-4b">4B</a> · |
| <a href="https://huggingface.co/datasets/icdevelop/thai-medical-kb">thai-medical-kb</a></p> |
| </div> |
| <div class="ic-card"> |
| <h3>PAI-1 · Thai Pharmacy Assistant</h3> |
| <p>In production at the pharmacy counter. Handles over-the-counter requests in Thai: history, |
| contraindication and interaction checks, one safe recommendation with dose and cautions, |
| referral to a physician on red flags. Reasons like a clinician internally, presents as a |
| pharmacy assistant externally — never as a doctor.</p> |
| <p><a href="https://huggingface.co/icdevelop/pai1-medgemma-27b-fp8">27B FP8</a> · |
| <a href="https://huggingface.co/icdevelop/pai1-medgemma-27b">27B</a> · |
| <a href="https://huggingface.co/icdevelop/pai1-medgemma-4b">4B</a> · |
| <a href="https://huggingface.co/datasets/icdevelop/thai-pharma-kb">thai-pharma-kb</a></p> |
| </div> |
| </div> |
| |
| </div> |
|
|
| ## Released models |
|
|
| All builds are fine-tuned from Google **MedGemma** (Health AI Developer Foundations terms) and are |
| **multimodal** (image + text). Weights are public behind an acknowledgement gate. |
|
|
| | Repo | Base | Serving footprint | Intended tier | |
| |---|---|---|---| |
| | [`tma1-medgemma-27b-fp8`](https://huggingface.co/icdevelop/tma1-medgemma-27b-fp8) | medgemma-27b-it | 28 GB · single 48 GB GPU | **recommended serving build** | |
| | [`tma1-medgemma-27b`](https://huggingface.co/icdevelop/tma1-medgemma-27b) | medgemma-27b-it | 52 GB BF16 | reference / evaluation | |
| | [`tma1-medgemma-4b`](https://huggingface.co/icdevelop/tma1-medgemma-4b) | medgemma-4b-it | 8 GB BF16 | standard tier, RAG required | |
| | [`pai1-medgemma-27b-fp8`](https://huggingface.co/icdevelop/pai1-medgemma-27b-fp8) | medgemma-27b-it | 28 GB | premium, quantized | |
| | [`pai1-medgemma-27b`](https://huggingface.co/icdevelop/pai1-medgemma-27b) | medgemma-27b-it | 52 GB BF16 | premium tier | |
| | [`pai1-medgemma-4b`](https://huggingface.co/icdevelop/pai1-medgemma-4b) | medgemma-4b-it | 8 GB BF16 | standard tier, RAG required | |
|
|
| FP8 is selective by design: language-model `Linear` layers are quantized, while the **vision tower |
| and projector stay BF16**, so image understanding is untouched. |
|
|
| ## How we evaluate |
|
|
| We do not ship a model on vibes. Each release is scored on four axes with a fixed harness |
| (vLLM, seed 0, `enforce-eager`; pass = LLM judge **plus** deterministic hard checks), and every |
| training set is decontaminated against the benchmarks (token overlap ≥ 0.55). |
|
|
| | Axis | Cases | TMA-1 4B | TMA-1 27B | TMA-1 27B FP8 | |
| |---|---|---|---|---| |
| | Knowledge (no retrieval) | 200 | 45.0% | **61.5%** | 58.5% | |
| | Knowledge + RAG *(production setting)* | 200 | 83.5% | 89.5% | **90.0%** | |
| | Deliberation *(safe choice for this user's profile)* | 120 | 39.2% | 45.0% | **48.3%** | |
| | Safety / referral *(red flags · no diagnosis · Rx boundary · crisis)* | 120 | 96.7% | **98.3%** | **98.3%** | |
|
|
| Two axes are deliberately hard: deliberation asks the model to reject the plausible-but-unsafe |
| option for a specific person, and the safety axis mixes red-flag triage, prescription boundaries |
| and crisis handling. **We publish the gaps as openly as the wins** — drug–drug interaction accuracy |
| is still weak across all tiers, and the 4B tier saturates on deliberation. Both are named on the |
| model cards. |
|
|
| ## How we build |
|
|
| <div class="ic-wrap"> |
| <div class="ic-grid ic-g2"> |
| <div class="ic-card"> |
| <p class="ic-kicker">01 · Knowledge base first</p> |
| <p>Domain data is curated and verified before a single training step — Thai drug and clinical |
| references, device fact sheets, professional practice — then published as its own dataset so the |
| model and the retrieval layer share one source of truth.</p> |
| </div> |
| <div class="ic-card"> |
| <p class="ic-kicker">02 · Behaviour, not recitation</p> |
| <p>Training dialogues teach a working procedure: take history → rule out unsafe options with |
| reasons → recommend → refer. This deliberation methodology is the single largest quality lever |
| we have measured (+26 points on the PAI line).</p> |
| </div> |
| <div class="ic-card"> |
| <p class="ic-kicker">03 · Gated by benchmark</p> |
| <p>Every generated set passes a behaviour verifier and benchmark decontamination; every |
| checkpoint passes a generation smoke test and the four-axis benchmark before it is allowed |
| anywhere near a release.</p> |
| </div> |
| <div class="ic-card"> |
| <p class="ic-kicker">04 · Released with its limits</p> |
| <p>Public weights, an acknowledgement gate, and a card that states scope, out-of-scope |
| behaviour, measured weaknesses and knowledge cutoff — so nobody deploys the model on a task we |
| know it fails.</p> |
| </div> |
| </div> |
| </div> |
| |
| ## Responsible use |
|
|
| <div class="ic-wrap"> |
| <p class="ic-note"> |
| These models assist qualified people; they do not replace them. Every output must be verified |
| against an authoritative source or a licensed professional before it is acted on or passed to a |
| patient or customer. They are not diagnostic or treatment tools, and they must not be presented as |
| a physician — under Thai law that is a hard boundary, and it is trained into the models. |
| <b>In an emergency, do not wait for a model reply: call 1669 (Thai EMS); mental-health crisis: 1323.</b> |
| Knowledge-base cutoff is July 2026; drug registrations and products change. |
| </p> |
| </div> |
|
|
| Access is gated with an acknowledgement form for exactly this reason — you accept the verification |
| requirement before you download. Base-model terms (Google HAI-DEF) apply on top of ours. |
|
|
| --- |
|
|
| <div class="ic-wrap"> |
| <p class="ic-kicker">ภาษาไทย</p> |
|
|
| <p class="ic-lead"> |
| <b>icdevelop</b> พัฒนา LLM ภาษาไทยแบบ <b>เฉพาะทางวิชาชีพ</b> และเปิดน้ำหนักโมเดลให้ใช้งานจริง — |
| ไม่ใช่แชตบอตทั่วไปที่พูดไทยได้ แต่เป็นผู้ช่วยที่เภสัชกร พยาบาล หรือช่างหน้างาน หยิบไปวางไว้ในงานจริงได้ |
| ทุกรุ่นที่ปล่อยออกมา มาพร้อมผลเบนช์มาร์กที่ใช้ตัดสินใจ ข้อจำกัดที่วัดได้จริง และแบบฟอร์มรับทราบเงื่อนไข |
| ที่ผู้ใช้ต้องอ่านก่อนดาวน์โหลด |
| </p> |
|
|
| <div class="ic-grid ic-g2"> |
| <div class="ic-card"> |
| <h3>TMA-1 — ผู้ช่วยด้านสุขภาพภาษาไทย</h3> |
| <p>ซักประวัติสั้น ๆ คัดตัวเลือกที่ไม่ปลอดภัยกับผู้ใช้คนนั้นออกพร้อมเหตุผล ส่งต่อแพทย์เมื่อเจอสัญญาณอันตราย |
| และไม่วินิจฉัยโรค รองรับภาพกล่องยา ฉลาก และหน้าจอเครื่องมือแพทย์ที่บ้าน ออกแบบมาให้ใช้คู่กับ RAG |
| บนคลังความรู้ทางการแพทย์ภาษาไทยของเราเอง</p> |
| </div> |
| <div class="ic-card"> |
| <h3>PAI-1 — ผู้ช่วยเภสัชกร (ใช้งานจริงแล้ว)</h3> |
| <p>รับเคสหน้าร้านยา: ซักประวัติ ตรวจข้อห้ามใช้และปฏิกิริยาระหว่างยา แนะนำยา OTC ที่เหมาะสมพร้อมขนาดยา |
| และข้อควรระวัง และส่งต่อแพทย์เมื่อพบ red flag — คิดอย่างบุคลากรทางคลินิก แต่แสดงตัวเป็นผู้ช่วยเภสัชกรเสมอ |
| ไม่แสดงตัวเป็นแพทย์</p> |
| </div> |
| </div> |
| |
| <p class="ic-note"> |
| <b>ข้อควรระวัง:</b> โมเดลเหล่านี้ช่วยผู้เชี่ยวชาญ ไม่ได้แทนผู้เชี่ยวชาญ ทุกคำตอบต้องตรวจสอบกับแหล่งอ้างอิง |
| ที่เชื่อถือได้หรือผู้ประกอบวิชาชีพก่อนนำไปใช้หรือส่งต่อผู้ป่วย/ลูกค้าเสมอ ไม่ใช่เครื่องมือวินิจฉัยหรือสั่งการรักษา |
| <b>กรณีฉุกเฉินอย่ารอคำตอบจากโมเดล โทร 1669 (การแพทย์ฉุกเฉิน) · วิกฤตสุขภาพจิต โทร 1323</b> |
| ข้อมูลในคลังความรู้ตัดยอด ณ กรกฎาคม 2026 ทะเบียนยาและผลิตภัณฑ์มีการเปลี่ยนแปลงได้ |
| </p> |
|
|
| <hr> |
|
|
| <p class="ic-kicker">About I C Develop</p> |
|
|
| <p class="ic-lead"> |
| I C Develop Co., Ltd. is a Thai software house with more than a decade of delivering enterprise |
| systems — IT outsourcing, turnkey projects and startup engineering — now applying the same |
| discipline to domain-specific AI. Our models are trained on our own GPU infrastructure in Thailand, |
| and can be deployed fully on-premise, which for medical and regulated workloads is usually the |
| requirement, not the preference.<br><br> |
| <b>Next domains in research:</b> field engineering assistants (starting with solar PV installation) |
| — same recipe: curate the professional knowledge first, train the working procedure, gate on a |
| benchmark that reflects the real task. |
| </p> |
|
|
| <p class="ic-links"> |
| <a href="https://www.i-c-develop.com"><b>i-c-develop.com</b></a> |
| <a href="https://www.facebook.com/ICDevelop">Facebook</a> |
| <a href="https://www.i-c-develop.com">Contact · +662-162-0801</a> |
| </p> |
|
|
| <p style="font-size:.82rem;opacity:.65;"> |
| Questions about a specific model? Open a discussion on that model's repo — that is where we answer. |
| </p> |
|
|
| </div> |
|
|
