README.md

---
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>