Save time every day
- Find a document, a procedure, or a letter in seconds.
- Prepare quotes, reports, meeting notes, and replies.
- An AI installed on your premises: your data stays in.
The founder
Jason Weber
Founder of JAZBOND · AI consultant and custom technical projects
Aircraft pilot, then international consultant: mining in Madagascar, aviation, maritime, and setting up companies in Africa and the Indian Ocean. Today I help businesses save time with AI on their everyday work, without their data ever leaving the building. And when the need goes beyond the office, I design the custom technical solution: sensors, connected devices, applications, machines.
- Aviation
- Mining
- Maritime
- Connected devices
- Private AI
- Africa and the Indian Ocean
Background
Why this background matters to your project.
Twenty years in fields where an approximation is paid for in cash: the cockpit, Madagascar's mining projects, maritime transport, business set-ups across Africa.
Regulated, multicultural environments where you do not trust a system because it looks like it works. You verify it, you log it, and you verify it again.
That is precisely the discipline enterprise AI lacks. A model that is wrong is wrong with poise, in flawless prose. Safeguards are not improvised afterwards: they are designed together with the system.
Milestones
The path, in a few dates.
1992 – 2002
Professional pilot: business aviation and bush flying.
2006 – 2016
International consultant, above all in Madagascar's mining sector: the Soalala iron ore mine, bauxite and mineral sands at Fort-Dauphin. Also aviation, maritime, and setting up companies in Africa and the Indian Ocean.
2010 – 2016
Drones: design, reconnaissance, and surveying in Madagascar.
2022 – 2023
Advisory work on launching a shipping company in the Indian Ocean.
Since 2023
Founder of JAZBOND; four years of training in generative AI.
Since late 2024
Armonitia: design, launch, and operation.
Conviction
AI gives time back. It replaces no one.
That is a position, and it carries technical consequences — not just good intentions.
I do not design systems that decide in place of your teams. I design systems that take away what nobody wants to do: re-keying the same data into three tools, hunting twenty minutes for a document, rewriting the same report every month. The time given back returns to the work where your organisation actually creates value.
This is not a sales promise, it is an architectural constraint. Actions that commit the organisation go through human validation, roles and permissions are explicit, and the log states who decided what. A system built that way cannot stand in for the person: it prepares the work, the person decides.
There is a practical reason on top of the right one: a tool teams experience as a threat does not get used. Adoption is settled there, not in the demo.
In practice
What I do for a business.
A custom technical project
- Sensors and connected devices: get an alert before a breakdown.
- Complete web and mobile apps.
- A working prototype before committing a big budget.
The proof
Armonitia, carried through end to end.
Before proposing an architecture to an organisation, I carried one through to the end on my own account.
Armonitia is a platform in production: web, App Store, Google Play, seven languages, working payments, monitoring, metrics and backups. Designed, built, deployed and maintained single-handed.
- 24
- specialised agents, each with its domain and voice
- 528
- named, characterised sub-agents, 22 per guardian
- 50,413
- routing keywords, about 95 per sub-agent
- 7 languages
- across three platforms: web, iOS, Android
Routing you can audit
Vector proximity backed by a keyword safety net. The machine does not choose at random, and you can check afterwards why it chose what it chose.
Answers anchored in a corpus
Document retrieval before every answer: the agent leans on identified sources instead of the model's own memory. That is what separates an assistant that cites from one that invents.
A decision graph, not a black box
Language, distress, intent, retrieval, drafting, review: every step is named, observable and separately testable. When behaviour goes wrong, you know which step to blame.
Knowledge the system maintains
Raw sources are never modified. The synthesis stays reconstructible from them, and the source is what carries authority. The same mechanism transfers to an organisation's own expertise.
All of it opens in one click.
What transfers to your organisation is the architecture, not the content. The hard part is never the model: it is knowing the business, shaping that knowledge, and building the controls that stop the machine inventing.
Transfer
Four industries, one method.
A well-designed agent system carries from one field to another, provided the business is modelled first.
DECARBOSHIP, maritime decarbonisation
Wind and solar modules in a 45-foot container, telescopic masts, retractable hydrojets, LFP storage. The technical model reproduces the engineer's four reference cases to within a percent per tonne.
GreenREE Lab, green chemistry
Bio-sourced coumarin from sweet vernal grass, as an additive for iron-cobalt-nickel electrodeposition. Protocol backed by a peer-reviewed literature review.
Drone anti-jamming, aviation
Resistance to RF jamming and instrument landing using sub-gigahertz LoRa beacons, without GPS. Navigation, frequencies, and what you do when the signal disappears.
HOLIVET, animal health
75 supplements, 74 conditions, 17 breeds, diagnosis crossing symptoms with conditions, weight-based dosage. The same architecture as Armonitia, in an entirely different field.
These are studio prototypes and studies. They are not here to pad a list: they show what actually transfers from one field to another.
Method
What ships can be verified.
Every fix I ship comes with a test that fails when the defect returns. And I check that the test really fails before I believe it.
A measurement, never an impression. I record what did not work as carefully as what did: on an AI system, that is the only way to know what can be promised to a client, and what cannot.