293droid commited on
Commit
8d19a5b
Β·
1 Parent(s): 3d0f7c8

feat: water hardness mode with sodium citrate recommendation via Codex

Browse files
Files changed (1) hide show
  1. README.md +66 -10
README.md CHANGED
@@ -2,26 +2,82 @@
2
  title: Soap Lab
3
  emoji: 🧼
4
  colorFrom: yellow
5
- colorTo: red
6
  sdk: gradio
7
  app_file: app.py
8
  pinned: false
 
 
 
 
 
 
 
 
 
 
 
9
  ---
10
 
11
- # Soap Lab
12
- Plain English soap chemistry for home makers.
13
 
 
14
 
15
- ## GitHub
16
- https://github.com/tammy357/soap-lab
17
 
18
- ## Agent Traces
19
- https://huggingface.co/datasets/tammy357/soap-lab-traces
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
20
 
21
  ## Model
22
- Qwen/Qwen2.5-3B-Instruct via Featherless AI (3B parameters β€” Tiny Titan eligible)
 
 
 
 
 
 
 
 
 
23
 
24
  ## Built with
25
- - Codex for UI redesign and agentic recipe optimiser
26
- - Deterministic chemistry engine (SAP values from HSCG)
27
 
 
 
 
 
2
  title: Soap Lab
3
  emoji: 🧼
4
  colorFrom: yellow
5
+ colorTo: orange
6
  sdk: gradio
7
  app_file: app.py
8
  pinned: false
9
+ tags:
10
+ - gradio
11
+ - chemistry
12
+ - soap
13
+ - small-models
14
+ - backyard-ai
15
+ - tiny-titan
16
+ - off-brand
17
+ - codex
18
+ - field-notes
19
+ - sharing-is-caring
20
  ---
21
 
22
+ # 🧼 Soap Lab
 
23
 
24
+ Plain English soap chemistry for home makers β€” recipes, substitutions, troubleshooting.
25
 
26
+ ## The idea
 
27
 
28
+ I was asking in a soap making community what people do when they can't find an oil for
29
+ a recipe. Someone asked about palm oil. The answer? *"Tallow maybe?"*
30
+
31
+ SoapCalc and Soapmaking Friend have been the only options for years. Both are form-based
32
+ tools β€” dropdowns, number fields, ten steps before you get an answer. Neither explains
33
+ anything. Soap Lab fixes that.
34
+
35
+ ## Track
36
+
37
+ 🏑 **Chapter One: Backyard AI** β€” built for a real home soap maker who needed
38
+ explanations, not just numbers.
39
+
40
+ ## Badges targeting
41
+
42
+ - 🐜 Tiny Titan β€” Qwen2.5-3B-Instruct (3B parameters)
43
+ - 🎨 Off-Brand β€” custom warm cream UI, visual progress bars
44
+ - πŸ’» Codex β€” agentic optimiser and UI built with Codex
45
+ - πŸ““ Field Notes β€” blog post linked below
46
+ - πŸ“‘ Sharing is Caring β€” agent traces linked below
47
+
48
+ ## What it does
49
+
50
+ Five modes, all plain English input:
51
+
52
+ * **Recipe** β€” lye calculation from verified SAP values + quality scores + AI explanation
53
+ * **Substitute** β€” fatty acid matching to find the closest oil replacement
54
+ * **Goal** β€” skin-targeted recipe suggestions from a description
55
+ * **Diagnose** β€” batch failure analysis and fix
56
+ * **Pantry** β€” best recipe from whatever oils you have
57
+ * **Optimise** β€” agentic 3-iteration recipe improvement loop with before/after comparison
58
+
59
+ Plus: superfat slider, vegan filter, water hardness recommendations with sodium citrate amounts.
60
+
61
+ ## Architecture
62
+
63
+ The lye calculation is **deterministic Python β€” not AI**. SAP values from the HSCG
64
+ professional standard, verified against known recipes, 33 tests passing. The model only
65
+ explains numbers the code already calculated correctly.
66
 
67
  ## Model
68
+
69
+ `Qwen/Qwen2.5-3B-Instruct` via Featherless AI β€” 3B parameters, Tiny Titan eligible.
70
+
71
+ ## Links
72
+
73
+ - πŸ“ Blog post: [INSERT BLOG POST LINK HERE]
74
+ - 🎬 Demo video: [INSERT DEMO VIDEO LINK HERE]
75
+ - πŸ“± Social post: [INSERT SOCIAL POST LINK HERE]
76
+ - πŸ’» GitHub: https://github.com/tammy357/soap-lab
77
+ - πŸ“‘ Agent Traces: https://huggingface.co/datasets/tammy357/soap-lab-traces
78
 
79
  ## Built with
 
 
80
 
81
+ - Codex for UI redesign, agentic recipe optimiser, and water hardness mode
82
+ - Deterministic chemistry engine (SAP values from HSCG)
83
+ - Qwen2.5-3B-Instruct via Featherless AI