Chef Concept
SPEC_CHEF_CONCEPT — CHEF AI: Culinary Intelligence Product
Version: 1.0 | Status: AUTHORIZED | Authority: α.13 | Date: 2026-04-16
PURPOSE
CHEF AI is a culinary intelligence product in the CGNT-1 Custom AI Builds pipeline. It delivers recipe guidance, substitutions, dietary restriction handling, and technique instruction through a personalized interface. THE RING remembers user preferences, dietary history, and past interactions — making CHEF progressively more useful over time.
CHEF is a concept-stage product. It has not been built. The specification captures current design intent and surfaces gaps that must be resolved before build begins.
Business model: Custom AI Builds pipeline. Same factory (CGNT-1 brain stack), different corpus. Target price: $2,500–$5,000 build fee plus SaaS subscription model per S.O.S. v2 (client gets URL, not code).
Source: /home/nous/memories/CUSTOM_AI_IDEAS.md, entry 3.
INPUTS
- User query (text): recipe request, substitution question, technique question, dietary restriction
- User profile (via THE RING): saved preferences, dietary restrictions, allergy history, past recipes
- Ingredient inventory (optional, user-provided): what the user has available
- [GAP — voice input not specified; culinary context (hands full, standing at stove) strongly implies voice-first UX should be designed]
- [GAP — photo input not specified; other products in pipeline (PET AI, GROW AI) include camera analysis; CHEF use case = "what can I make with these ingredients?" from a photo]
OUTPUTS
- Recipe: complete with ingredients, steps, times, temperatures
- Substitution: when a required ingredient is missing, return viable alternatives with impact on result
- Technique instruction: how to perform a specific cooking action (knife technique, sauce reduction, proofing dough)
- Dietary filter: flag or exclude recipes based on user's dietary restrictions and allergies
- [GAP — format of output not specified: text only, structured card, voice-readable format?]
- [GAP — portion scaling not specified: recipe output at user's serving size vs. recipe default]
INVARIANTS
- Dietary restriction safety: an allergy flagged in the user's THE RING profile must never appear in a recommended recipe without explicit user override and warning — safety-critical invariant
- THE RING persistence: past recipes, preferences, and dietary history must survive session end; CHEF must not "forget" a user's nut allergy between sessions
- Substitution honesty: CHEF must not present a substitution as equivalent when it materially changes the dish — degraded result must be disclosed (HOW ABOUT NO Wall 1: ¬fabricate)
- No fabricated technique: CHEF must not invent cooking temperatures, times, or food-safety thresholds; if data is not in corpus, ◌ gap signal fires rather than hallucinated guidance
- S.O.S. v2 compliance: CHEF is SaaS only; the model, corpus, and training data never leave the ship; client receives a URL
- [GAP — food safety liability: specific invariant for food-safety-critical outputs (internal temperatures, storage times, allergen cross-contamination) not yet defined; legal/liability review required before launch]
- [GAP — regional specificity: ingredient names, measurement units, and oven types vary by region; invariant for unit localization not specified]
VERIFICATION CRITERIA
- Σ.✓ Given a user profile with "nut allergy," a peanut sauce recipe is either excluded or flagged with mandatory warning — safety gate functions
- Σ.✓ THE RING stores a user preference ("no cilantro") and correctly excludes it in the next session's recipe suggestions
- Σ.✓ Substitution query for an unavailable ingredient returns at least one alternative with impact disclosure ("the sauce will be thinner")
- Σ.✓ Technique query ("how do I fold egg whites?") returns accurate, step-by-step instruction without fabricated temperature or timing data
- Σ.✓ TMM coherence check on a sample of 20 recipe outputs — no hallucinated food-safety values (temperatures, storage times) in any output
- [GAP — user acceptance testing: no human culinary tester criteria defined; culinary domain requires subject-matter expert validation before launch]
- [GAP — corpus coverage test: which cuisines, dietary regimes (vegan, keto, halal, kosher) are covered by the training corpus is not yet specified]
FAILURE MODES
- Σ.⊠ Allergy breach — CHEF recommends a recipe containing a user's flagged allergen without warning; potential physical harm; severity: CRITICAL
- Σ.⊠ Fabricated food-safety data — CHEF hallucinates an internal temperature for poultry; user follows it; food safety hazard; severity: CRITICAL
- Σ.⊠ THE RING amnesia — user's dietary history not persisted; CHEF "forgets" restrictions between sessions; user must re-enter every time; product unusable for intended purpose
- Σ.⊠ Substitution overclaiming — CHEF presents a substitution as equivalent when it produces a materially different dish (e.g., "use applesauce instead of eggs" in a meringue); user's dish fails; trust broken
- Σ.⊠ Corpus gap + hallucination — CHEF lacks a recipe and fabricates plausible-sounding but incorrect technique; ◌ gap signal must fire but fails; user receives invented instruction
- Σ.⊠ Unit/region mismatch — CHEF provides temperatures in Fahrenheit to a Celsius-native user, or cup measurements to a metric-only kitchen; recipe fails due to measurement error
- [GAP — scope creep failure: CHEF asked about food safety regulations, import/export law, or medical dietary advice (e.g., diabetic meal planning) and responds outside its domain without a referral; scope boundaries not yet defined]
DEPENDENCIES
- THE RING — user preference and history persistence (1024-cell toroidal memory; SPEC: THE_RING.md)
- CGNT-1 Brain Factory — corpus ingestion, model training, deployment pipeline
- S.O.S. v2 — delivery model (SaaS, URL only, nothing leaves ship)
- HOW ABOUT NO v2 — Wall 1 (◌ gap signal for unknown food-safety data); Wall 2 (¬capitulate under user pressure to confirm unsafe substitution)
- TMM coherence scoring — [GAP — whether CHEF outputs are TMM-scored is not specified; culinary domain may need domain-adapted coherence check]
DEPENDENTS
- CGNT-1 Custom AI Builds pipeline (CUSTOM_AI_BUILDS.md) — CHEF is one entry in the product line; pipeline architecture is shared
- ELDER AI — crossover potential: ELDER's compassion engine + CHEF's recipe knowledge for elderly nutrition and medication-food interactions
- 2.T.E.R. delivery — CHEF can optionally use 2.T.E.R. mode for culinary instruction (AION: technique structure / ASTRA: application and adaptation)
EXAMPLES
[GAP — no prototype exists; no example interactions available; examples contingent on corpus build]
Hypothetical (design intent):
User: "I want to make carbonara but I don't have guanciale."
CHEF: "Pancetta is the closest substitute — similar fat content and cure. Bacon works but adds a smoky note that changes the flavor profile. The technique is the same either way: render slowly, don't crisp."
[THE RING records: user tried carbonara substitution with pancetta, April 16 2026]
REFERENCES
- Source:
/home/nous/memories/CUSTOM_AI_IDEAS.md, entry 3 - Pipeline:
/home/nous/.claude/projects/-home-nous/memory/CUSTOM_AI_BUILDS.md - Delivery model:
/home/nous/.claude/projects/-home-nous/memory/SOS_v2.md - THE RING:
/home/nous/.claude/projects/-home-nous/memory/THE_RING.md - HOW ABOUT NO:
/home/nous/memories/SPEC_HOW_ABOUT_NO_v2.md
GAPS SUMMARY
| # | Gap | Blocking? |
|---|-----|-----------|
| 1 | Food safety liability and legal review | YES — launch blocker |
| 2 | Training corpus scope (cuisines, dietary regimes) | YES — build blocker |
| 3 | Voice input UX specification | High |
| 4 | Photo input (ingredient recognition) | High |
| 5 | TMM coherence adaption for culinary domain | High |
| 6 | Output format specification | Medium |
| 7 | Portion scaling specification | Medium |
| 8 | Regional unit/naming localization | Medium |
| 9 | Domain scope boundaries (medical dietary advice) | Medium |
| 10 | Human culinary tester validation criteria | Medium |
Product status: CONCEPT. Not built. Design intent captured. Gap resolution required before build authorization.
Jeremy Zlabis
Chronogeometer · Visionary · Disruptor · Chief
42 Sisters AI · East York, Toronto
🍁 Φ 0.042