Intelligent Cooktop Safety

Catch dangerous cooking
conditions early.

HomeHalo uses thermal sensing and on-device intelligence to monitor cookware temperatures, recognize developing hazards, and alert users before a dangerous situation escalates.

See the Prototype → Request a Pilot See how it works
Live Thermal View
Cooktop monitoring — active
247°C
Front Left
Empty pan — hazard
182°C
Front Right
Boil risk rising
94°C
Rear Left
Simmering — normal
22°C
Rear Right
Inactive
⚠️
Alert: Front left burner unattended 14 min — empty pan detected. Sending notification.
~50%
of home fires involve cooking equipment (NFPA)
#1
behavioral cause: unattended cooking
<5 min
empty pan can reach ignition temperature
100%
privacy-preserving — no cameras, no video
How It Works

Four layers of intelligent protection

From sensing to intervention, HomeHalo operates as a continuous, context-aware safety system — not just a sensor.

01
🔭

Infrared Sensing

A compact thermal sensor mounts above your cooktop — no cameras, no video. Continuously reads surface heat at the cookware level.

02
📊

Thermal Analysis

Proprietary algorithms analyze temperature trajectories, heating rate, and spatial gradients — far beyond simple threshold detection.

03
🧠

Context Inference

A probabilistic engine interprets cooking state: is the stove attended? Is there liquid in the pan? Is temperature accelerating abnormally?

04
🔔

Progressive Alert

From gentle reminders to push alerts to optional power shutoff — responses scale to hazard severity and user-defined preferences.

Features

Precision safety, thoughtfully designed

🌡️

Empty-Pan Detection

Identifies rapid monotonic temperature rise characteristic of cookware with no liquid or food mass — a leading precursor to oil ignition events.

💧

Boil-Over Prediction

Monitors convective thermal signatures and edge temperature patterns to detect impending boil-over before it reaches the heating element.

⏱️

Unattended Cooking Detection

Infers user presence and cooking supervision through cookware displacement, utensil interaction events, and prolonged inactivity patterns.

🔒

Privacy First

Thermal imaging inherently obscures personal visual details. No RGB camera. No video recording. Coarse spatial heat maps only — processed locally.

🔧

Retrofit Compatible

Designed to operate independently of stove electronics. Works with any existing gas, electric, or induction cooktop — no installation professional required.

📱

Smart Notifications

Push alerts to your phone when hazards are detected. Configurable sensitivity, quiet hours, and adaptive thresholds that learn your cooking habits.

Hardware

Designed to fit anywhere

HomeHalo ships in the GU10 light bulb form factor — the global standard for kitchen range hoods, recessed ceiling lighting, and accent fixtures. Zero new wiring. Instant retrofit.

💡
GU10 Form Factor

Drop-in replacement for existing GU10 fixtures. No rewiring, no new hardware, no contractor required.

🍳

Kitchen Range Hood Lighting

Most range hoods already use GU10 bulbs positioned directly above the cooktop — the perfect vantage point for continuous thermal monitoring.

💫

Recessed Ceiling Lighting

Standard kitchen recessed downlights are GU10-compatible. HomeHalo installs in seconds, replacing an existing bulb with zero visible change.

🏨

Accent & Track Lighting

GU10 track and accent fixtures provide flexible coverage for studio kitchens, extended-stay suites, and open-plan cooking areas.

Smart Home

Works with the ecosystem you already use

HomeHalo integrates natively with leading smart home platforms — no separate hub, no extra apps, no friction.

Apple HomeKit
Amazon Alexa Amazon Alexa
Google Home
Siri Shortcuts

Ask Siri to check your stove, set automations in Home, or get Alexa to announce an alert in every room. HomeHalo speaks the language of your home.

💡
Built into the GU10 light bulb form factor. Smart home intelligence is baked directly into the same fixture you already have above your cooktop — range hood, recessed ceiling, or accent lighting. One bulb. Zero extra hardware.
Hospitality & Commercial

Trusted by leading hospitality brands

Extended-stay and studio-kitchen properties are an ideal deployment environment — high unit counts, consistent GU10 fixture use, and meaningful fire risk reduction at scale.

Active Partnership
Studio 6

Studio 6 by G6 Hospitality

Secured initial deployment order covering 102 extended-stay studio units. Studio 6's kitchenette format is an exact fit for HomeHalo's GU10 range-hood form factor.

G6 Hospitality LLC — 102 units contracted

In Discussions
Hilton LivSmart Studios

LivSmart Studios by Hilton

Active conversations underway with Hilton's LivSmart Studios brand — a growing portfolio of extended-stay properties with studio kitchens across North America.

hilton.com/livsmart-studios ↗

In Discussions
Element Hotels by Marriott

Element Hotels by Marriott

Discussions ongoing with Element Hotels, Marriott's sustainability-focused extended-stay brand known for full kitchen suites and health-conscious design standards.

element-hotels.marriott.com ↗

Hardware Partnership

Hardware Partner — Phase 2

HomeHalo's Phase 2 GU10 design is being explored in collaboration with Cosmo Products, LLC.

Cosmo Appliances

Cosmo Products, LLC

Integrated hardware development partnership with Cosmo Products, LLC — HomeHalo is exploring hardware development with Cosmo Products, LLC for the Phase 2 GU10 form factor — bringing manufacturing expertise and product integration knowledge to the project.

How It Works

The intelligence behind HomeHalo

No machine learning black boxes. Every detection is a clear physical rule — written in code, measurable, tunable, and explainable. Built and tested by Aki Suda using real kitchen recordings.

1

Find the Pot

Every pixel ≥ 60°C is grouped into thermal "blobs." The largest blob is the vessel.

2

Find the Liquid

Inside the pot, Otsu's method splits the cooler rim (~75°C) from the hotter liquid surface (~85–92°C).

3

Watch Over Time

Track how fast temperature rises (dT/dt), how fast area changes (dA/dt), and how full the pot is (fill ratio).

4

Apply the Rules

Physics-based rules detect hazard states — from boiling detection to spillover confirmation to smoke plume recognition.

Detection Algorithms

Physics-based detection — measurable, tunable, explainable

🔥
Critical Alert

Empty Pan Heating

An empty vessel heats like a rocket — nothing inside absorbs energy. Detected by rapid uniform temperature rise with no cool-rim / hot-liquid structure inside.

dT/dt ≥ 3°C/sec for ≥ 5 seconds
AND no liquid structure inside
OR average temp ≥ 190°C
♨️
Critical Alert

Burning Utensil

The simplest rule. If the cookware's average temperature crosses 180°C, the alarm fires immediately — no other condition needed.

vessel average temp ≥ 180°C
→ alert fires unconditionally
Can co-fire with empty pan
💧
Critical Alert

Spillover (3-Stage)

A boil-over follows a 3-stage story. Boiling must be confirmed first, then the pot fills to 85% capacity, then heat appears outside the vessel — cooler than the liquid inside.

Stage 1: boiling confirmed
Stage 2: fill ratio ≥ 0.85, rising
Stage 3: dA/dt ≥ 10 px²/s outside,
new pixels ≥ 4°C cooler than liquid
🫧
Warning

Boiling Detected

Boiling water mixes itself — all pixels read nearly the same temperature. Hot food being cooked is patchy. This one physical difference separates the two reliably.

liquid temp: 84–101°C
surface spread (std) ≤ 3°C
AND: stable 10s OR fill rising
🥘
Warning

Sauce Burning Risk

A watery sauce can never exceed ~100°C while water remains. When the same vessel exceeds 105°C and keeps climbing, the water has evaporated — burning is imminent.

Act 1: 75–100°C for ≥ 5 seconds
Act 2: temp > 105°C AND
dT/dt ≥ 0.3°C/sec, still climbing
🍳
Warning

Partial Burning

A small piece of food stuck to the pan can burn while the rest cooks normally. Detected as a persistent super-hot minority patch inside an otherwise normal vessel.

hot spot ≥ 160°C, ≥ 8 pixels
spot < 40% of vessel area
persists ≥ 2 seconds
💨
Warning

Smoke Detection

Smoke is semi-transparent to thermal cameras — it appears as a flickering warm patch. Steam from boiling water is excluded because its heat source never reaches 140°C.

source nearby ≥ 140°C
plume pixels 35–58°C
pixel flicker std ≥ 2°C/8 frames
patch ≥ 60px for ≥ 1 second
⚡
Warning

Spillover Risk

Issued before spillover occurs — when boiling liquid has risen to fill 85% or more of the vessel and is still climbing. Gives the user time to act before the spill.

boiling confirmed
fill ratio ≥ 0.85 AND rising
state held for 8 seconds
(survives the fill-ratio dip at spill)
⚡
Detected

Sudden Reaction

A water droplet hitting a hot pan causes a local temperature shock over in under a second. Distinguished from stirring by duration — the same pattern lasting over 1.5 seconds means a person is cooking.

local ΔT ≥ 20°C between frames
rate ≥ 30°C/sec, area 4–400px
duration ≤ 1.2s = reaction
duration ≥ 1.5s = human stirring
👨‍🍳
Informational

Human Involvement

Stirring or flipping food repeatedly swaps the cooler top layer with the hotter bottom. Detected as sustained local thermal shock. Shown in green — never triggers an alarm.

same sudden-reaction pattern
repeating at same spot ≥ 1.5s
→ re-labelled HUMAN_INVOLVEMENT
shown green, not an alert
Privacy Architecture

No cameras. No video. No cloud.

🌡️

Thermal Only

The sensor produces a 192×256 grid of temperature numbers — not a photograph. Faces, bodies, and personal belongings are completely invisible. The camera literally cannot see what you look like.

💻

On-Device Processing

Every detection algorithm runs locally on the Raspberry Pi 4. Temperature data never leaves the device. Safety decisions happen in milliseconds with no internet dependency.

🔇

No Recording

HomeHalo does not store video, images, or continuous thermal data. Event logs record only detection timestamps and thermal statistics — never a reconstruction of your kitchen activity.

Software Architecture

Four files. One system.

The HomeHalo codebase is built for clarity — each file has exactly one job, and the live camera and the replay tool share the exact same detection brain.

🧠 thermal_hierarchy.py The Brain

All 10 detection rules live here. Finds the pot, finds the liquid inside it, tracks both over time, applies every rule, and maintains the full lifetime history of every thermal object. Includes a 20-case self-test suite.

👁️ thermal_monitor.py The Eyes (Live)

Connects to the Thermal Master P2 via USB-C. Reads 256×386 frames at 25fps, validates each frame, converts raw 16-bit values to °C (value ÷ 64 − 273.15), and feeds the brain in real time. Runs headless on Raspberry Pi.

▶️ thermal_playback.py The Replay Tool

Feeds recorded or simulated frames through the identical brain — so any tuning done in playback behaves exactly the same live. Includes a physics simulator with ready-made scenarios: boiling, spillover, empty pan, frying overheat.

🛠️ thermal_core.py The Toolbox

Color palettes, drawing utilities, event log writer (events.jsonl), and session metrics counter. Also contains the original v1 detector kept for comparison benchmarking.

System Output
📄 events.jsonl — live event log
[t=18.0s] BOILING detected
→ core_mean: 88.4°C, std: 1.8°C
→ fill_ratio: 0.71, rising
[t=23.4s] SPILLOVER_RISK armed
→ fill_ratio: 0.87, dA/dt: 42px²/s
[t=24.1s] SPILLOVER confirmed
→ footprint_dAdt: 385px²/s
→ escaped pixels 6.2°C cooler
📄 lifetimes.jsonl — blob histories
track_id: 1, kind: vessel
lifetime_s: 24.1, frames: 302
phases: initial→heating→boiling
temp_min: 76.0°C, max: 92.6°C
area_min: 2716px, max: 5755px
peak_dAdt: 514.5 px²/s
Tuned Parameters

Every value measured from real data

These aren't guesses — they were measured from real kitchen recordings and refined through field testing. Values marked (was X) show what changed from initial estimates.

🫧 Boiling Detection

BOIL_BAND_LOW_C84°CCamera reads boiling water low due to surface foam and angle — measured from real recording
BOIL_BAND_HIGH_C101°CUpper bound of liquid boiling range
EVEN_SURFACE_STD3.0°CBoiling water: 1.5–2.5°C spread. Cooking food: 4–7°C. Wall between them
BOIL_FILL_RATIO0.35Was 0.55 — lowered because tilted camera sees smaller liquid surface
BOIL_MIN_DURATION10 secTemperature must be stable within 2°C for this duration for sign (a)

💧 Spillover Detection

SPILL_RISK_FILL0.85Fill ratio at which spillover risk is armed — 85% of pot area covered by rising liquid
SPILL_RISK_HOLD_S8.0 secMemory survives this long — fill ratio drops at moment of actual spill
SPILL_MIN_DADT10 px²/sReal recorded spill: ~385 px²/s. Gate of 10 eliminates single-frame false spikes
SPILL_COOLER_DELTA4.0°CEscaped liquid must read at least 4°C cooler than liquid inside
SATELLITE_MIN_AREA100 pxWas 12 — raised to reject splash specks, keep real escaped puddles

🔥 Empty Pan & Burning

EMPTY_PAN_DTDT3°C/secHeating rate threshold — empty pans heat much faster than loaded cookware
EMPTY_PAN_MIN_RAMP5 secMust sustain this rate for 5 seconds — eliminates brief thermal spikes
EMPTY_PAN_ABS_TEMP190°CBackup: a uniform pan already at 190°C fires immediately regardless of rate
BURNING_UTENSIL_C180°CCookware average at or above this = burning. No other condition required.

⚡ Reaction vs. Stirring

REACTION_MIN_DELTA20°CWas 8°C — raised to eliminate small wobbles, only big shocks count
REACTION_DTDT30°C/secWas 20°C/sec — raised for same reason, eliminates slow thermal drift
REACTION_MAX_LIFE1.2 secUnder this = sudden reaction (droplet splash). Over 1.5s = human stirring
STIRRING_MIN_S1.5 secThe time wall between a single shock and repeated human cooking activity
Live Demo

See HomeHalo in action

Phase 1 prototype test footage — Thermal Master P2 sensor mounted above a real cooktop, running live thermal analysis on the Raspberry Pi 4.

Test 1 — Thermal Monitoring · Live Sensor Feed Phase 1 Prototype · Raspberry Pi 4 + Thermal Master P2

Test 1 — Sensor Feed

Test 2 — Heat Detection

Test 3 — Temp Trajectory

Test 4 — Edge Inference

🌡️

Real Thermal Data

All footage is live sensor output from the Thermal Master P2 — 256×192 IR resolution at 25 Hz, capturing real cooktop heat patterns in real time.

💻

On-Device Processing

The Raspberry Pi 4 runs all thermal trajectory analysis locally. No cloud. No lag. Safety decisions happen on-device in milliseconds.

🔬

Phase 1 Validation

These tests validate the sensing and inference pipeline before miniaturization into the GU10 form factor for Phase 2 deployment.

Phase 1 — Working Prototype

Built from scratch.
Tested on a real cooktop.

Raspberry Pi 4 — HomeHalo edge processing unit
🖥 RASPBERRY PI 4 — EDGE PROCESSING UNIT
Raspberry Pi 4 — I/O ports and GPIO
🔌 I/O & GPIO
Arduino Nano + thermal sensor — early enclosure prototype
🖨 3D-PRINTED ENCLOSURE

Raspberry Pi 4 + Thermal Master P2

The Phase 1 prototype uses a Raspberry Pi 4 as the on-device edge processing unit paired with the Thermal Master P2 thermal imaging sensor. All inference runs locally — no cloud dependency, no data leaving the device.

An early Arduino Nano prototype in a custom 3D-printed enclosure was used to validate sensor placement and field-of-view geometry above a real cooktop.

Sensor: Thermal Master P2 — 256×192 IR, ±1.5°C
Processor: Raspberry Pi 4 — edge inference
Early prototype: Arduino Nano + thermal array
Enclosure: Custom 3D-printed, designed in-house
Software: Python — thermal trajectory analysis
Nvidia Jetson
PROCESSING
Nvidia Jetson
Edge AI board — all analysis on-device.
Thermal Master P2
THERMAL SENSING
Thermal Master P2
IR array — no camera, no video.
Pi NoIR Camera V2
DEVELOPMENT REFERENCE
Pi NoIR Camera V2
Used for field-of-view calibration only — not for detection. raspberrypi.com ↗
Phase 2 — Under Development

Moving from prototype
to deployable product.

The Phase 2 design is planned to fit into a standard GU10 range-hood socket — the same fitting used by bulbs in most range hood exhaust systems. The goal is to make installation as simple as replacing a light bulb, with no new wiring and no modifications to the exhaust system itself.

💡
GU10 Form Factor
Designed to fit the GU10 socket already present in most range hood exhaust systems. No modifications, no new wiring — integrates with existing exhaust infrastructure.
🔔
Smart Alerts
Planned push notification system with progressive alerts — a reminder first, then a phone alert. Smart-home integrations (HomeKit, Alexa, Google Home) in scope for Phase 2.
🏭
Production Enclosure
Production-grade form factor being explored with hardware partner Cosmo Products, LLC. Optional appliance shutoff for plug-in cooktops also under consideration.
🔌
Designed to integrate with existing exhaust systems
Most range hoods already have a GU10 bulb socket built in. The Phase 2 HomeHalo sensor is planned to occupy that socket — no drilling, no electrician, no new infrastructure. The exhaust fan continues to work exactly as before.
Development Status

Serious, working technology.
Moving toward deployment.

HomeHalo is a validated prototype, not a finished product. Here is an honest account of where it stands and where it is going.

WORKING TODAY
Phase 1 — Validated Prototype
✓
Infrared thermal array sensing above a live cooktop
✓
On-device processing on a Raspberry Pi 4 — fully local, no cloud
✓
Temperature trajectory analysis (dT/dt, area change, fill ratio)
✓
Physics-based hazard detection — empty pan, boiling, overheating
✓
Real cooktop testing, iterated with fire-safety professional feedback
✓
U.S. Provisional Patent filed · Published in Silconeer Magazine
UNDER DEVELOPMENT
Phase 2 — Deployable Product
→
GU10 form-factor miniaturization (fits range-hood socket, no wiring)
→
Smart-home integrations — Apple HomeKit, Alexa, Google Home
→
Push notifications and progressive alert system
→
Production-grade enclosure with hardware partner Cosmo Products, LLC
→
Optional appliance shutoff integration for plug-in cooktops
→
Pilot deployments — exploring extended-stay hospitality and senior living
🔒
Privacy by design
No RGB camera. No video. No images. HomeHalo captures thermal data only. Core thermal analysis is designed to run locally on-device.
In the Field

Developed with feedback from fire-safety professionals.

Aki presented early HomeHalo prototypes and the underlying problem to members of the San Ramon Valley Fire Protection District, and incorporated their feedback into later iterations of the system.

Aki with SRVFD Captain
SRVFD Captain — Station 34
Engine 34 crew
Engine 34 — SRVFD
SRVFD Station
San Ramon Valley FPD
Aki Suda, Founder of HomeHalo
Our Story

Built with purpose.
Driven by impact.

I'm Aki Suda, a 17-year-old junior at Brave Christian High School in Dublin, CA. I built HomeHalo because I believe the technology to detect dangerous cooking conditions already exists — it just hasn't been put in the right place yet.

The idea started simply: most cooking fires happen because someone walks away. Smoke detectors only respond after a fire starts. I wanted something that could recognize a developing hazard before smoke ever forms — the moment a pan overheats or a burner is left unattended.

I built the first prototype with an Arduino Nano and a thermal imaging sensor in a 3D-printed enclosure. I tested it above my own stove. I presented early prototypes to members of the San Ramon Valley Fire Protection District to understand the problem better — and their feedback shaped the direction of the project.

1
U.S. Provisional Patent filed
Real
Cooktop tests with SRVFD feedback
Phase 2
GU10 miniaturization underway

"The technology exists. The problem is real. I built something to prove it."

— Aki Suda, Founder
Contact

Interested in a
pilot or partnership?
Let's talk.

HomeHalo is a working prototype actively moving toward a deployable product. If you're a researcher, safety organization, hospitality operator, or anyone interested in evaluating the technology — reach out.

✉️
Get in touch
info@homehalo.ai
🏨
Hospitality & multi-unit deployments welcome
📦
Technology shared openly — Aki welcomes questions and conversations
Send a Message
WHERE IT STANDS
Phase 1
Working prototype — validated on real cooktops
Provisional patent filed · Silconeer Magazine (Aug 2026)
→ Hardware partner: Cosmo Products, LLC (Phase 2)
→ Exploring pilot deployments in safety-critical settings
🔥
Why it matters
158,400 home cooking fires per year in the US. Unattended cooking is the #1 cause. HomeHalo detects it before it starts.
Live R&D Log

The Build Log

Live R&D, hardware iterations, testing, and progress from the bench. 13 entries across 12 months — from first Arduino experiments to Jetson AI inference, firefighter feedback, and YC applications.

Read the Build Log →