README.md

metadata

title: Soap Lab
emoji: 🧼
colorFrom: yellow
colorTo: green
sdk: gradio
app_file: app.py
pinned: false
tags:
  - gradio
  - chemistry
  - soap
  - small-models
  - backyard-ai
  - tiny-titan
  - off-brand
  - codex
  - field-notes
  - sharing-is-caring
  - track:backyard
  - sponsor:openai
  - achievement:offgrid
  - achievement:offbrand
  - achievement:sharing
  - achievement:fieldnotes

🧼 Soap Lab

Plain English soap chemistry for home makers — recipes, substitutions, troubleshooting.

The idea

I was asking in a soap making community what people do when they can't find an oil for a recipe. Someone asked about palm oil. The answer? "Tallow maybe?"

SoapCalc and Soapmaking Friend have been the only options for years. Both are form-based tools — dropdowns, number fields, ten steps before you get an answer. Neither explains anything. Soap Lab fixes that.

Track

🏡 Chapter One: Backyard AI — built for a real home soap maker who needed explanations, not just numbers.

Badges targeting

• 🐜 Tiny Titan — Qwen2.5-3B-Instruct (3B parameters)
• 🎨 Off-Brand — custom warm cream UI, visual progress bars
• 💻 Codex — agentic optimiser and UI built with Codex
• 📓 Field Notes — blog post linked below
• 📡 Sharing is Caring — agent traces linked below

What it does

Five modes, all plain English input:

• Recipe — lye calculation from verified SAP values + quality scores + AI explanation
• Substitute — fatty acid matching to find the closest oil replacement
• Goal — skin-targeted recipe suggestions from a description
• Diagnose — batch failure analysis and fix
• Pantry — best recipe from whatever oils you have
• Optimise — agentic 3-iteration recipe improvement loop with before/after comparison

Plus: superfat slider, vegan filter, water hardness recommendations with sodium citrate amounts.

Architecture

The lye calculation is deterministic Python — not AI. SAP values from the HSCG professional standard, verified against known recipes, 33 tests passing. The model only explains numbers the code already calculated correctly.

Model

Qwen/Qwen2.5-3B-Instruct via Featherless AI — 3B parameters, Tiny Titan eligible.

Links

• 📝 Blog post: https://huggingface.co/blog/build-small-hackathon/soap-lab-tallow-maybe
• 🎬 Demo video: https://youtu.be/10OhPgMqAKo
• 📱 Social post: https://x.com/flower56778/status/2066095227429961777
• 💻 GitHub: https://github.com/tammy357/soap-lab
• 📡 Agent Traces: https://huggingface.co/datasets/tammy357/soap-lab-traces

Built with

• Codex for UI redesign, agentic recipe optimiser, and water hardness mode
• Deterministic chemistry engine (SAP values from HSCG)
• Qwen2.5-3B-Instruct via Featherless AI

🧼 Soap Lab

Plain English soap chemistry for home makers — recipes, substitutions, troubleshooting.

The idea

I was asking in a soap making community what people do when they can't find an oil for a recipe. Someone asked about palm oil. The answer? "Tallow maybe?"

SoapCalc and Soapmaking Friend have been the only options for years. Both are form-based tools — dropdowns, number fields, ten steps before you get an answer. Neither explains anything. Soap Lab fixes that.

Track

🏡 Chapter One: Backyard AI — built for a real home soap maker who needed explanations, not just numbers.

Badges targeting

• 🐜 Tiny Titan — Qwen2.5-3B-Instruct (3B parameters)
• 🎨 Off-Brand — custom warm cream UI, visual progress bars
• 💻 Codex — agentic optimiser and UI built with Codex
• 📓 Field Notes — blog post linked below
• 📡 Sharing is Caring — agent traces linked below

What it does

Five modes, all plain English input:

• Recipe — lye calculation from verified SAP values + quality scores + AI explanation
• Substitute — fatty acid matching to find the closest oil replacement
• Goal — skin-targeted recipe suggestions from a description
• Diagnose — batch failure analysis and fix
• Pantry — best recipe from whatever oils you have
• Optimise — agentic 3-iteration recipe improvement loop with before/after comparison

Plus: superfat slider, vegan filter, water hardness recommendations with sodium citrate amounts.

Architecture

The lye calculation is deterministic Python — not AI. SAP values from the HSCG professional standard, verified against known recipes, 33 tests passing. The model only explains numbers the code already calculated correctly.

Model

Qwen/Qwen2.5-3B-Instruct via Featherless AI — 3B parameters, Tiny Titan eligible.

Links

• 📝 Blog post: https://huggingface.co/blog/build-small-hackathon/soap-lab-tallow-maybe
• 🎬 Demo video: https://youtu.be/10OhPgMqAKo
• 📱 Social post: https://x.com/flower56778/status/2066095227429961777
• 💻 GitHub: https://github.com/tammy357/soap-lab
• 📡 Agent Traces: https://huggingface.co/datasets/tammy357/soap-lab-traces

Built with

• Codex for UI redesign, agentic recipe optimiser, and water hardness mode
• Deterministic chemistry engine (SAP values from HSCG)
• Qwen2.5-3B-Instruct via Featherless AI