Multilingual by design
Locale, script and direction are modelled in the data, not translated on top of it.
About us
We research, engineer and operate AI software.
HublinkTech develops AI-native platforms for SMEs and enterprise operations across multiple industries. We engineer the same systems for organisations solving problems of their own. Founded on 23 years in global business, built as an AI innovation technology company.
How we build
What we hold to.
AI engineered in, not added on
Retrieval, agents and generation run as services with boundaries and evaluation, not features bolted on.
We operate what we build
The team that engineers a system runs it in production. That is what keeps the standards honest.
Research and innovation
The problems we work on.
The platforms depend on systems that no API call solves on its own: they have to run multi-tenant, in production, and in more than one language. Each item below is open work, and each is hard for a specific reason.
Multilingual NLP
Our own interfaces ship in four languages, and Ara is being built to answer in five, including right-to-left scripts. Model quality does not transfer between languages, so classification, extraction and generation are evaluated per language instead of being assumed to carry over from English.
Retrieval that can be checked
Ara is being built to answer from current tariff schedules, rules of origin and corridor requirements rather than from search results, and to attach its sources. The hard part is staying grounded while the underlying regulation changes — and building a system that is allowed to say it does not know.
Agents with boundaries
Agents deployed against real processes — quoting, triage, classification — need defined tools, permitted actions and explicit failure paths. Where an autonomous step ends and a human one begins is a design decision we make again for every process.
Matching beyond keywords
Hublinkly pairs buyers and suppliers on industry, geography, capacity and compliance posture rather than keyword overlap. Scoring entities against one another means working with sparse, uneven signals and producing a ranking that can be traced back to its inputs.
Trade and regulatory data
HS classification, duties across origin and destination pairs, preferential rates, documentation requirements and route risk come from sources that share no schema and change without notice. Turning them into structured data — with confidence scores and a path for the cases that cannot be resolved — is continuous work.
Evaluation as part of the system
Every deployment carries an evaluation set and a human baseline, and ships only if it beats that baseline. Building those sets for multilingual, regulated, judgement-heavy tasks, where the correct answer is itself contested, is part of the research.
Who leads the work
Babak Abedi
Founder | Director, HublinkTech
HublinkTech is an AI software company. Its founder, Babak Abedi, is a computer engineering graduate who directs its engineering and spent 23 years in international B2B trade — which is why the problems the platforms attack are understood in detail rather than assumed: matching, classification, documentation, compliance. Solving them is research and engineering.
The platforms came first. Hublinkly, Clavix360 and Ara exist because the problems they solve were met first-hand in cross-border trade — and that is where the research question started, not where it ended. Knowing a problem in detail tells you what to build; it does not tell you how to make a system solve it reliably across languages, regulations and tenants. That part is researched, engineered and measured.
Where we are deep
What we know well.
Platform engineering
Multi-tenant systems, production engineering and the infrastructure underneath.
Applied AI
Agents, retrieval and multilingual NLP deployed against real processes.
Data engineering
Pipelines, warehousing and lineage from source system to decision.
Trade and commerce
Two decades of cross-border B2B, encoded into the products.
Our footprint
A global footprint.
Istanbul
- Istanbul
What drives us
The principles we ship by.
Systems, not deliverables
We are measured by whether the thing runs, not whether it shipped.
Production shape from day one
Multi-tenant, observable and tested from the first commit, not hardened later.
Evidence over assertion
Numbers that trace to a source, models measured against a baseline.
How we work
Small team, deep involvement.
HublinkTech is founder-led and product-first. The people who design a system are the people who build and operate it, which is why we take on fewer things and go further into each one.
In one document
The company presentation
The same story as this page, as a fifteen-slide deck: the platforms, the technology, the business model and the roadmap. Readable in the browser, downloadable in English or Turkish.
Where we work
Built in one place. Deployed everywhere.
Everything we build is designed, built and operated from Istanbul. The platforms themselves run wherever our users do.
