Visual AI — designed and operated in France
Horama turns your image streams — drones, cameras, field captures — into reliable operational decisions. Specialized, frugal, explainable models, deployed close to the field, even offline.
01 — Manifesto
Horama is a French sovereign visual-AI company. We turn image streams into operational decisions: models each trained for one mission, deployed close to the field, where every decision can be proven against the image.
Technology proven in demanding environments, from defence to industrial applications — now in production across six sectors.
Of the technology stack from French sources
Sectors with systems in production
Of processing possible locally, on site
Dependency on external servers
02 — Approach
90% of our technology stack originates from French sources. When components come from foreign sources, we isolate them in closed environments under exclusively French management. We deploy locally — on your infrastructure or your site — or on French cloud providers. At no point does your data depend on servers outside our control.
We do not use large general-purpose models that burn considerable resources. We design small, sober models, sized for their task. It is an ecological choice as much as an economic one — and it is what makes local processing possible, right where the work happens.
AI is not meant to be generic: its value comes from specialization to a precise business use. Each of our models is trained for one mission — reliability and relevance where a universal model would stay approximate.
Every decision our tools produce can be explained. The operator who relies on our systems daily can access the proof behind what the AI asserts — every indicator can be traced back to the image that grounds it. That is the condition of trust.
03 — Sectors
Our models run in production, in the field, for real clients. For confidentiality, we describe missions — never references.
AI-powered supervision of large infrastructure projects. Seeing the site in real time means acting before the problem: fewer delays, more safety, funding under control.
A — The system
Drones, fixed cameras and mobile captures, paired with ruggedized stations installed on site: the entire analysis runs locally, even with no connectivity. Your video streams never leave the site — only a selected fraction of the data reaches an analysis component hosted in France.
B — The proof
Every observation is dated, geolocated and attached to its visual source. Nothing is declarative: every indicator can be traced back to the image that grounds it. Reports constitute enforceable proof of progress — admissible to your funders, your partners and the authorities. You no longer steer on statements, but on verifiable facts.
C — The supervision
Progress — actual state versus schedule, deviation by work package and by zone. Safety — continuous monitoring of protective equipment and site rules. Financial compliance — cross-checking orders, deliveries and observed execution. Anticipation — early detection of foreseeable delays and supply tensions.
Typical case — 01
Several work fronts at once — runway, terminal, taxiways — over several years. Every zone supervised in parallel, delays detected live, progress proven at every funding milestone.
Typical case — 02
An isolated site, often in poorly connected mountain terrain, with high risks for crews. Continuous supervision, even offline, and proof of progress for the lenders.
Visual audits of venues and travelling events: the location, the security arrangements already in place and the management of crowd flows, assessed from the image. Built with a specialist in event security.
A — The site audit
Nexus performs the visual audit of a venue or travelling event: the layout of the location and the security arrangements already in place, assessed point by point from the image — an objective, documented, shareable inventory.
B — The flows
The analysis covers flow management: circulation axes, density zones, control points. An objective reading that informs the decisions of security teams — humans decide, the image documents.
Helios inspects photovoltaic plants by drone: thermography and multi-sensor imaging detect anomalies panel by panel, before they weigh on production.
A — The detection
The drone survey covers the plant row by row. Thermography reveals thermal anomalies invisible to the naked eye — every panel inspected, every defect located and documented.
B — The qualification
Multi-sensor imaging crosses spectra to qualify each detected anomaly. Operations teams intervene where it is useful — before the defect weighs on production.
Auris analyses driver behaviour and expressions in real time: adapting the vehicle environment — temperature, alerts — and strengthening on-board safety, up to the detection of carjacking situations.
A — The analysis
Driver behaviour and expressions are analysed in real time, inside the cabin: vigilance, fatigue, attention. A continuous, explainable reading of the actual state at the wheel.
B — The adaptation
The vehicle adjusts its environment to the driver's state — temperature, alerts — and detects abnormal situations, up to carjacking. Visual analysis becomes a safety feature.
Terra surveys plantations by drone to assess the sanitary state of crops and the water stress of soils — in service of more precise, more resilient agriculture.
A — Crop health
The drone survey assesses the sanitary state of crops at plot scale. What ground surveys take weeks to show, one imaging campaign reveals — and documents.
B — Water stress
Image analysis reveals the water stress of soils and guides irrigation decisions. More precise, more resilient agriculture — grounded in observation, not estimation.
Visual recognition that adapts the content offered to each visitor in real time — an eight-year-old does not receive the same message as a corporate group. Already in service in parks in France and Portugal.
A — The recognition
Visual recognition adapts the content pushed to each visitor in real time: the right message, at the right moment, to the right person. A specialized AI serving the experience — not the other way round.
B — In production
Prism runs today in parks in France and Portugal, integrated into the venues' mobile applications. Proof that a specialized AI belongs in mainstream visitor experiences.
04 — Differentiation
05 — International
Fast-growing emerging markets have an immense need for infrastructure — and for the intelligent supervision that makes it reliable. Pharos is the spearhead of this expansion: built for demanding terrain, operating offline, sovereign by design.
First CES in Las Vegas. Use cases validated, first qualified international contacts.
Systems in production across six sectors, in France and Portugal.
CES Las Vegas — going international. Bringing field-proven technology to emerging markets, Pharos first.
06 — Team
Horama designs AI visual-analysis systems deployed in demanding environments, from defence to industrial applications.
07 — Frequently asked questions
Sovereign visual AI is image-analysis artificial intelligence whose technology and data remain under national control. At Horama, 90% of the technology stack originates from French sources; foreign components are isolated in closed environments under exclusively French management, and deployment happens on-premise or on French cloud infrastructure.
As close to the field as possible. Our systems process footage locally, on your infrastructure or on ruggedized stations installed on site. When hosting is required, it runs on French cloud providers. You never depend on external servers outside French control — on a Pharos site, video streams never leave the premises.
Because the value of AI comes from specialization. We design small, sober models, each trained for one precise business mission: more reliable than a general-purpose model that stays approximate, far less energy-hungry — and that frugality is what makes local processing possible, with no remote datacenter.
Every observation produced by Pharos is dated, geolocated and attached to its visual source. Nothing is declarative: every indicator can be traced back to the image that grounds it. Reports constitute enforceable proof of progress, admissible to funders, partners and authorities.
Yes. The ruggedized stations installed on site run the entire processing pipeline locally: analysis continues even with no connectivity. It is a design requirement, built for isolated or poorly connected terrain — mountain sites, sprawling facilities, constrained environments.
No, it is not a prerequisite. We handle the complete installation of the required capture devices, working with specialized partners. If you already have equipment, we connect it and use it as is, with no replacement. Horama coordinates everything, from installation to the delivery of analyses.
Frugal AI is a deliberately small, sober model trained for one precise task rather than for everything. Compared with large general-purpose models, it consumes far less energy and resources, is cheaper to run, and fits on embedded hardware — which makes local, on-site processing possible with no remote datacenter. Both an ecological and an economic choice.
Explainable AI lets you understand and justify every decision it produces. In computer vision, every detection can be traced back to the image that grounds it. The operator always has access to this visual proof — the condition of trust, and a prerequisite for regulated sectors and enforceable uses such as proof of construction progress.
Yes, compliance is built into the design. Local processing, French data control and the explainability of every decision meet GDPR and EU AI Act requirements. Streams do not leave the site when unnecessary, and hosting, when required, runs on French cloud providers — with no dependency on servers outside the European Union.
Six sectors: building and large construction (Pharos — supervision, proof of progress), events and security (Atlas and Nexus — visual audit, crowd-flow management), energy (Helios — solar farm drone inspection and thermography), mobility (Auris — driver analysis), agriculture (Terra — crop health and water stress by drone) and leisure and culture (Prism — visual recognition for parks). Each sector relies on a specialized model.
A project starts with a study of the terrain and constraints, then installation of the capture devices, commissioning of the platform and team training, and finally continuous supervision. Cost depends on scale: on a large infrastructure project, the solution typically represents a marginal fraction of the overall budget. Email contact@horama.ai for a tailored estimate.
