Olfactory AI · Digital Nose (e-nose) HQ Daejeon, Korea · Research Lab Seoul · iliasai.com

Robots can see, hear and touch. We give them a sense of smell. ILias AI builds digital olfaction: a multichannel olfactory sensor array paired with deep learning that recognizes an odor's full chemical fingerprint, on the bench and on quadruped robots that patrol, search and detect.

WHY SMELL IS THE MISSING SENSE

Robots and autonomous systems increasingly work where people work, and many of the most dangerous hazards in plants, facilities and public spaces are invisible to cameras: toxic gas leaks, chemical releases, contraband. A robot that can smell can detect a hazard, identify what it is, and move toward its source before people are exposed. Olfaction is also a largely untapped data modality for multimodal AI.

TECHNOLOGY

Multichannel olfactory sensor array + AI deep learning (not spectroscopic): learns odor fingerprints, not just a single concentration value.

Odor library built on the desktop reference system.

Digital olfactory data platform: patented in Korea.

PRODUCTS

Desktop model: lab reference and odor-library builder. A thin pin-type probe, connected by hose, samples air from closed boxes through small gaps, without opening them.

Quadruped robot-mounted payload: compact unit for mobile patrol and gas detection.

Both work together for spatial search and gas detection.

APPLICATIONS BY INDUSTRY

Public safety & security Narcotics, explosives and toxic gas screening at customs and airports.

Industrial safety & energy Gas leak detection in chemical plants, battery and chip fabs.

Robotics Quadruped robot-mounted patrol of hazardous areas.

Mobility & autonomous vehicles Cabin air quality, and early warning of battery thermal runaway or gas leaks.

Aerospace & space habitats Cabin air and life-support monitoring in sealed environments.

Smart buildings & data centers Indoor air, overheating and gas leak monitoring for facilities.

Healthcare & wellness Breath and odor biomarkers (research stage).

Food & fragrance Freshness, quality and scent checks for consumer goods.

AI & digital olfaction data TSM odor data as a new sensory modality for multimodal AI.

* These are areas where our technology can apply; some are at the research or exploration stage.

R&D: 3D ODOR MAPPING FOR ROBOT ACTIVE SEARCH

Mapping the odor fingerprint in 3D (not just concentration), with uncertainty, to estimate source location and steer a robot's active search. First focus: toxic gas industrial safety.

CREDENTIALS

Korean Patent No. 102912639: "Digital Olfactory Data Service Platform" (registered Jan 9, 2026.

Commercialization based on ETRI technology transfer Electronics and Telecommunications Research Institute, Korea).

Planned olfactory data center in the Ulsan/Busan region.

THE ASK

A short conversation, and a small pilot of our quadruped robot-mounted digital nose at an industrial site or another facility of your choice.

Bumsuk Ko, CEO iliasai.com

Email: [email protected]

ILias AI · Olfactory AI / Digital Nose iliasai.com

Vision Paper · Olfactory AI · Digital Nose

Companion to the ILias AI one-pager · iliasai.com

I L i a s A I · TSM V IS ION

Turning smell into data. Building the standard with TSM. Color became data with RGB. Sound became portable with MP3. ILias AI is building TSM — the Transformer Scent Model — so that smell can become a shared, machine-readable standard.

1 · Vision — smell is the last sense without a digital standard

Once a sense has a shared coordinate system, the world can store it, search it, compare it, send it and teach it to machines. Color became data when everyone agreed on RGB; sound became something we stream and share when it was encoded in formats such as MP3.

Color RGB

Sound MP3

Smell TSM ?

Smell never got that moment. There is no widely adopted standard for capturing, describing and exchanging measured odor as data. People describe the same smell in different words, instruments respond differently and datasets use incompatible labels — so odor knowledge stays locked in noses, trained dogs and isolated lab instruments.

ILias AIʼs ultimate goal is to digitize human olfaction: a future where scents are searched like text, compared like images and generated like audio — with ILias AI as infrastructure of the olfactory AI stack.

2 · TSM — the Transformer Scent Model

TSM Transformer Scent Model) is the basic structure that defines an odor by linking its chemical formula, molecular structure, raw materials and the scent people perceive — together with the sensor fingerprint that measures it — through a transformer-based model into one digital odor record. Smell becomes data that can be stored, searched and exchanged: the coordinates of a smell, just as RGB values made color objective. ILias AIʼs goal is to propose TSM to international standards bodies so it becomes the standard data-storage technology for smell.

What TSM contains today (from company sources)

One digital odor record — chemical formula · molecular structure SMILES · raw material · perceived odor labels · sensor fingerprint, linked by one ID and embedding.

Bidirectional inference — forward: molecular structure SMILES → numeric scent labels such as “floralˮ or “earthy ;ˮ inverse: a target scent label → matching molecular structures, as data accumulates.

Four-step data pipeline — collect molecule–odor datasets → standard SMILES format → tokenize and vectorize into scent embeddings → train a transformer that defines the olfactory vector space.

Three-layer data schema — Assets Molecule_Info, Scent_Label) · Mapping Molecule_Scent_Map with intensity and confidence; Scent_Label_Map unifying external vocabularies) · Log & evolution Prediction_Log, Model_Info versions).

Sensor bridge — sensor response curves → substance identification → TSM coordinates and labels.

Toward an international odor-data standard

PROPOSED items for the standard submission

Odor fingerprint format for raw sensor- array signals

Measurement & calibration protocol — reference samples, temperature / humidity, drift

Labeling ontology — shared TSM reference vocabulary

Quality & uncertainty metadata on every record

Exchange API — query, compare and share odor records

Goal: propose TSM to international standards bodies

ILias AI · TSM Vision Paper · Oct 2026 Page 1 / 2

ILias AI · TSM VISION

3 · How ILias AI technology connects to TSM

SENSE

Multichannel olfactory sensor array

Desktop reference · quadruped robot-mounted

payload

›

ENCODE

TSM Odor fingerprint → vector

coordinates, labels, confidence

›

PLATFORM

Digital olfactory data platform

Patent 102912639 · odor library · planned data center

›

APPLY

Applications Safety · security · industry ·

robots

Sensor array + deep learning (not spectroscopic). Learns the full odor fingerprint, not a single concentration value — the raw signal TSM standardizes.

Desktop model = lab reference and odor-library builder. A thin pin-type probe on a hose samples air from closed boxes through small gaps without opening them, producing clean reference entries for the TSM library.

Quadruped robot-mounted payload = field data. Real backgrounds, humidity and mixtures teach TSM how odors behave outside the lab.

3D odor mapping R&D. Odor-fingerprint maps with uncertainty, source estimation and active search — TSM

gives every map point a common vocabulary.

Korean Patent No. 102912639, “Digital Olfactory Data Service Platformˮ (registered Jan 9, 2026 — the service layer that stores and delivers odor data; new library entries can be pushed to deployed devices over the cloud.

Commercialization based on ETRI technology transfer Electronics and Telecommunications Research Institute, Korea), and a planned olfactory data center in the Ulsan/Busan region.

4 · Imagining the future

VISION — an imagined future, not current products, customers or partnerships

Humanoid & service robots Humanoid and service robots carry a

TSM-compatible nose: they notice a leak, name it and walk toward its source.

Smart factories & industrial safety

Smart factories and industrial sites see invisible gas hazards as a

continuously updated map.

Customs, airports & security Shared odor libraries, updated over

the air, help screen for illicit substances.

Healthcare, food & environment

Freshness, quality and air are checked against reference fingerprints.

Digital media & communication

The aroma of bread and coffee travels with a family photo — odor exchanged like images today.

Standards & data infrastructure

TSM adopted through international standards bodies: scents searched like text, compared like images,

generated like audio.

“TSM aims to be a digital coordinate system that links the physical definition of a smell with how

people perceive it — so that every sensor, robot and lab can speak the same language of odor.ˮ

Letʼs build the standard for smell — together We invite robotics teams, AI labs, industrial-safety and research partners to co-develop TSM reference data, run pilots with our desktop and quadruped robot-mounted systems, and shape an open odor-data standard.

Bumsuk Ko CEO, ILias AI

iliasai.com

Email: [email protected]

Sources: ILias AI TSM deck & spec Apr 2026, IR Sep 2026. Section 4 is vision. Page 2 / 2

Olfactory AI · Digital Nose (e-nose) HQ Daejeon, Korea · Research Lab Seoul · iliasai.com

Robots can see, hear and touch. We give them a sense of smell. ILias AI builds digital olfaction: a multichannel olfactory sensor array paired with deep learning that recognizes an odor's full chemical fingerprint, on the bench and on quadruped robots that patrol, search and detect.

WHY SMELL IS THE MISSING SENSE

Robots and autonomous systems increasingly work where people work, and many of the most dangerous hazards in plants, facilities and public spaces are invisible to cameras: toxic gas leaks, chemical releases, contraband. A robot that can smell can detect a hazard, identify what it is, and move toward its source before people are exposed. Olfaction is also a largely untapped data modality for multimodal AI.

TECHNOLOGY

Multichannel olfactory sensor array + AI deep learning (not spectroscopic): learns odor fingerprints, not just a single concentration value.

Odor library built on the desktop reference system.

Digital olfactory data platform: patented in Korea.

PRODUCTS

Desktop model: lab reference and odor-library builder. A thin pin-type probe, connected by hose, samples air from closed boxes through small gaps, without opening them.

Quadruped robot-mounted payload: compact unit for mobile patrol and gas detection.

Both work together for spatial search and gas detection.

APPLICATIONS BY INDUSTRY

Public safety & security Narcotics, explosives and toxic gas screening at customs and airports.

Industrial safety & energy Gas leak detection in chemical plants, battery and chip fabs.

Robotics Quadruped robot-mounted patrol of hazardous areas.

Mobility & autonomous vehicles Cabin air quality, and early warning of battery thermal runaway or gas leaks.

Aerospace & space habitats Cabin air and life-support monitoring in sealed environments.

Smart buildings & data centers Indoor air, overheating and gas leak monitoring for facilities.

Healthcare & wellness Breath and odor biomarkers (research stage).

Food & fragrance Freshness, quality and scent checks for consumer goods.

AI & digital olfaction data TSM odor data as a new sensory modality for multimodal AI.

* These are areas where our technology can apply; some are at the research or exploration stage.

R&D: 3D ODOR MAPPING FOR ROBOT ACTIVE SEARCH

Mapping the odor fingerprint in 3D (not just concentration), with uncertainty, to estimate source location and steer a robot's active search. First focus: toxic gas industrial safety.

CREDENTIALS

Korean Patent No. 102912639: "Digital Olfactory Data Service Platform" (registered Jan 9, 2026.

Commercialization based on ETRI technology transfer Electronics and Telecommunications Research Institute, Korea).

Planned olfactory data center in the Ulsan/Busan region.

THE ASK

A short conversation, and a small pilot of our quadruped robot-mounted digital nose at an industrial site or another facility of your choice.

Bumsuk Ko, CEO iliasai.com

Email: [email protected]

ILias AI · Olfactory AI / Digital Nose iliasai.com

Vision Paper · Olfactory AI · Digital Nose

Companion to the ILias AI one-pager · iliasai.com

I L i a s A I · TSM V IS ION

Turning smell into data. Building the standard with TSM. Color became data with RGB. Sound became portable with MP3. ILias AI is building TSM — the Transformer Scent Model — so that smell can become a shared, machine-readable standard.

1 · Vision — smell is the last sense without a digital standard

Once a sense has a shared coordinate system, the world can store it, search it, compare it, send it and teach it to machines. Color became data when everyone agreed on RGB; sound became something we stream and share when it was encoded in formats such as MP3.

Color RGB

Sound MP3

Smell TSM ?

Smell never got that moment. There is no widely adopted standard for capturing, describing and exchanging measured odor as data. People describe the same smell in different words, instruments respond differently and datasets use incompatible labels — so odor knowledge stays locked in noses, trained dogs and isolated lab instruments.

ILias AIʼs ultimate goal is to digitize human olfaction: a future where scents are searched like text, compared like images and generated like audio — with ILias AI as infrastructure of the olfactory AI stack.

2 · TSM — the Transformer Scent Model

TSM Transformer Scent Model) is the basic structure that defines an odor by linking its chemical formula, molecular structure, raw materials and the scent people perceive — together with the sensor fingerprint that measures it — through a transformer-based model into one digital odor record. Smell becomes data that can be stored, searched and exchanged: the coordinates of a smell, just as RGB values made color objective. ILias AIʼs goal is to propose TSM to international standards bodies so it becomes the standard data-storage technology for smell.

What TSM contains today (from company sources)

One digital odor record — chemical formula · molecular structure SMILES · raw material · perceived odor labels · sensor fingerprint, linked by one ID and embedding.

Bidirectional inference — forward: molecular structure SMILES → numeric scent labels such as “floralˮ or “earthy ;ˮ inverse: a target scent label → matching molecular structures, as data accumulates.

Four-step data pipeline — collect molecule–odor datasets → standard SMILES format → tokenize and vectorize into scent embeddings → train a transformer that defines the olfactory vector space.

Three-layer data schema — Assets Molecule_Info, Scent_Label) · Mapping Molecule_Scent_Map with intensity and confidence; Scent_Label_Map unifying external vocabularies) · Log & evolution Prediction_Log, Model_Info versions).

Sensor bridge — sensor response curves → substance identification → TSM coordinates and labels.

Toward an international odor-data standard

PROPOSED items for the standard submission

Odor fingerprint format for raw sensor- array signals

Measurement & calibration protocol — reference samples, temperature / humidity, drift

Labeling ontology — shared TSM reference vocabulary

Quality & uncertainty metadata on every record

Exchange API — query, compare and share odor records

Goal: propose TSM to international standards bodies

ILias AI · TSM Vision Paper · Oct 2026 Page 1 / 2

ILias AI · TSM VISION

3 · How ILias AI technology connects to TSM

SENSE

Multichannel olfactory sensor array

Desktop reference · quadruped robot-mounted

payload

›

ENCODE

TSM Odor fingerprint → vector

coordinates, labels, confidence

›

PLATFORM

Digital olfactory data platform

Patent 102912639 · odor library · planned data center

›

APPLY

Applications Safety · security · industry ·

robots

Sensor array + deep learning (not spectroscopic). Learns the full odor fingerprint, not a single concentration value — the raw signal TSM standardizes.

Desktop model = lab reference and odor-library builder. A thin pin-type probe on a hose samples air from closed boxes through small gaps without opening them, producing clean reference entries for the TSM library.

Quadruped robot-mounted payload = field data. Real backgrounds, humidity and mixtures teach TSM how odors behave outside the lab.

3D odor mapping R&D. Odor-fingerprint maps with uncertainty, source estimation and active search — TSM

gives every map point a common vocabulary.

Korean Patent No. 102912639, “Digital Olfactory Data Service Platformˮ (registered Jan 9, 2026 — the service layer that stores and delivers odor data; new library entries can be pushed to deployed devices over the cloud.

Commercialization based on ETRI technology transfer Electronics and Telecommunications Research Institute, Korea), and a planned olfactory data center in the Ulsan/Busan region.

4 · Imagining the future

VISION — an imagined future, not current products, customers or partnerships

Humanoid & service robots Humanoid and service robots carry a

TSM-compatible nose: they notice a leak, name it and walk toward its source.

Smart factories & industrial safety

Smart factories and industrial sites see invisible gas hazards as a

continuously updated map.

Customs, airports & security Shared odor libraries, updated over

the air, help screen for illicit substances.

Healthcare, food & environment

Freshness, quality and air are checked against reference fingerprints.

Digital media & communication

The aroma of bread and coffee travels with a family photo — odor exchanged like images today.

Standards & data infrastructure

TSM adopted through international standards bodies: scents searched like text, compared like images,

generated like audio.

“TSM aims to be a digital coordinate system that links the physical definition of a smell with how

people perceive it — so that every sensor, robot and lab can speak the same language of odor.ˮ

Letʼs build the standard for smell — together We invite robotics teams, AI labs, industrial-safety and research partners to co-develop TSM reference data, run pilots with our desktop and quadruped robot-mounted systems, and shape an open odor-data standard.

Bumsuk Ko CEO, ILias AI

iliasai.com

Email: [email protected]

Sources: ILias AI TSM deck & spec Apr 2026, IR Sep 2026. Section 4 is vision. Page 2 / 2