Offline Edge Computing Reference Architecture for IoT Home Appliances
Pioneered by Yemek AI Engineering as a core standard of the APSNY Ecosystem.
NOTICE: This is not a commercial product. This is the open-source reference implementation of the HFSCA standard to accelerate manufacturers' proprietary edge-AI development. Manufacturers are not required to use this software, but it serves as the official architectural benchmark for FreeCuli certification.
Document Version: 2.1.0 (Production-Verified Deployment)
Author: FreeCuli / Yemek AI Engineering Team
License: GNU Affero General Public License v3.0 (AGPL-3.0)
This specification defines the production-ready micro-architectures constituting the Hands-Free Semantic Culinary Assistant (HFSCA) engine, the core technology powering the FreeCuli standard. In smart appliance manufacturing and distributed ambient computing environments (e.g., smart kitchens, connected ovens, display-integrated refrigerators), deploying continuous multimodal cloud infrastructure for local task execution creates an unsustainable financial burden. Cloud-based Large Language Model (LLM) processing loops and neural Text-to-Speech (TTS) rendering degrade linearly under multi-source ambient acoustic disruptions (e.g., kitchen exhaust fans, running water, blenders) and suffer from non-deterministic latency.
The FreeCuli HFSCA methodology completely decouples stateful computational culinary intelligence from the physical hardware execution layer. By operating as an asynchronous Edge-Computing runtime engine, the system structures monolithic recipe data into sequential streaming primitives (The Smart Playlist Matrix). Through deterministic on-device finite automata (DFA), client-side hardware abstractions (Screen Wake Lock API), memory-retention locks, localized data-pruning maps, and autonomous health-tech tracking layers, FreeCuli ensures 0-millisecond navigational latency, absolute network independence, and 0 USD server-side inference costs.
The FreeCuli reference platform isolates its execution footprint into four decoupled, independent modules. Commercial enterprises may consume these components as an integrated runtime ecosystem or implement individual modular layers inside existing embedded operating systems.
[USER INTERACTION: MONOLITHIC RECIPE DETECTED]
|
v
+-------------------------------------------------------+
| MODULE 3: HARDWARE PERSISTENCE |
| (Wake Lock Enforced & GC Lock) |
+-------------------------------------------------------+
|
v
+-------------------------------------------------------+
| MODULE 1: THE SMART PLAYLIST |
| (Deterministic State-Machine) |
+-------------------------------------------------------+
| |
[Volumetric Intent] [Acoustic Tokens]
v v
+-----------------------------+ +-----------------------------+
| MODULE 2: PRAGMATIC METRIC | | MODULE 1.2: ACOUSTIC IGNORE |
| CALIBRATOR (SUPABASE/ALG) | | SHIELD (IS_SPEAKING_REF) |
+-----------------------------+ +-----------------------------+
|
v
+-------------------------------------------------------+
| MODULE 4: AUTONOMOUS NUTRITIONAL PROFILING |
| (State Recovery & Gamification End-State) |
+-------------------------------------------------------+
The execution loop manages state transactions entirely on the device CPU, wrapping native browser or operating system Speech-to-Text (STT) abstractions without streaming raw audio data to remote clusters.
Continuous microphone capturing in high-reflectivity acoustic environments causes echo loops where the assistant re-captures its own spoken output. Forcing physical hardware interrupts causes driver-level audio clicking noises ("Bloop" transients).
HFSCA solves this without hardware interrupts via a software-level memory flag:
- Upon initiating speech synthesis, the execution loop locks
isSpeakingRef.current = true. - While true, the event handler for incoming voice frames intercepts the payload and executes an atomic
returninstruction, destroying the token data locally. - Upon completion, a 400ms atomic hardware settlement delay is enforced before resetting the flag, completely muting echo reflection artifacts.
Storing the entire global nutritional food database on embedded appliance hardware is impossible due to memory limits. HFSCA enforces a cloud-side data pruning protocol that maps ingredients into a normalized Gastronomic Density Table. The transformation uses the fundamental equation:
Mass (g) = Volume (ml) × Density Coefficient (ρ)
When a user sets a portion scaling attribute (m), the engine recalculates the structural text templates locally via an inline scale-processor (Rendered_Quantity = Base_Quantity × m).
HFSCA establishes an unbreakable session boundary by acquiring native hardware wake-state tokens (navigator.wakeLock.request('screen')). The visibility state safety hook ensures background tab recovery.
Mobile Chromium and WebKit browsers actively dump instantiated speech components from system memory during extended passive listening loops. HFSCA bypasses browser GC memory allocation routines by instantiating the text string array inside a global, immutable memory reference boundary (utteranceRef), preventing the browser's native memory garbage-collector from sweeping the active speaker stream.
Voice timers are executed on a global thread independent of the DOM unmount/remount cycles. If a user sets a 5-minute timer and navigates through multiple steps, the timer remains intact. When the timer expires, the assistant safely interrupts the current action, announces the alert, and executes a State Recovery Callback, autonomously re-reading the exact step the user was currently viewing, ensuring zero cognitive loss.
Upon reaching the final "Congratulations" execution state, the FreeCuli assistant ceases being a mere navigator and acts as an autonomous dietician. It extracts the scaled macronutrients (Protein, Fat, Carbs, Calories) of the active recipe and securely transacts them to the user's database footprint, incrementing a global gamification layer (total_meals_cooked) without requiring active user input.
This specification constitutes a detailed, public, and verifiable historical documentation of the HFSCA architecture. Under international patent frameworks (including USPTO Leahy-Smith America Invents Act "First-Inventor-to-File" rules, EPO Article 54 EPC, and WIPO guidelines), this document functions as Prior Art. Consequently, no hardware enterprise or corporate entity can legally file patent claims regarding "edge-computed volumetric-to-metric programmatic mapping linked to an on-device local acoustic state machine for cooking environments."
While HFSCA represents the open-source technical architecture, FreeCuli™ is the commercial "Ingredient Brand" (similar to Intel Inside) licensed to hardware OEMs.
When consumers see the "FreeCuli Inside" badge on a smart oven or refrigerator, it acts as a guarantee of 100% offline functionality, absolute data privacy, and zero-latency voice guidance.
The open-source community edition of this architecture is distributed under the strict GNU Affero General Public License v3.0 (AGPL-3.0). This ensures that any enhancements to the core technology remain open and available to the public.
For industrial manufacturers who require integrating the HFSCA architecture and complying with the FreeCuli standard within proprietary, closed-source hardware and embedded firmware, FreeCuli offers a Commercial B2B Certification and Trademark Agreement. This enterprise framework provides the manufacturer with:
- A formal Trademark & Certification License allowing them to legally display the
#FreeCULIcertification badge on their hardware and marketing materials. - Full access to official compliance test protocols, compliance verification tools, and technical whitepapers to ensure their custom edge architecture meets the global standard.