Senior Backend & AI Engineer

You must be physically located in Cyprus. This is a hard filter, not a preference. Applications from outside Cyprus will not be considered, regardless of relocation willingness on the candidate's side.

The hard part of AI sales is not making a model answer a prompt. It is giving the model the right customer, company, conversation, and market context at the right moment, then making sure the action it takes can be trusted.

You will own the spine that makes that possible: ingestion, identity, normalization, the Knowledge Graph, retrieval, orchestration, evaluation, and the production systems around them. The boundaries between backend engineering and applied AI are the work, not an org-chart problem to pass between teams.

Dija is small enough that architecture decisions still compound immediately. A reliable connector can remove hours from every sales team we serve. A bad retrieval decision can make an agent confidently wrong in every customer conversation. You will see both the user impact and the system trace.

This is not a research role where a demo is the finish line. You will ship production systems, instrument them, learn from failure, and decide which model or technique is actually worth the complexity. The right engineer here makes the whole product faster, safer, and more capable at once.

First 6 months (what you do Monday)

Take this seat if you want to work on the boundary where data quality, distributed systems, and model behaviour become one product. You want ownership of a meaningful surface area and a short path from a design decision to a customer outcome.

Do not take this seat if you want to spend your time only tuning prompts, only building CRUD endpoints, or writing architecture documents for another team to implement. If you need a large platform group to absorb the operational details, this is too early for you.

We will not ship a black box and call it intelligence. We will not accept unreliable ingestion because the demo looked good, or add a model because it is fashionable without measuring the behaviour it improves. We will make the smallest system that gives a customer a better sales outcome, then earn the right to make it bigger.

Show us a system you made reliable. Walk through the first failure, the evidence you collected, the trade-off you chose, and what changed after it reached real users. We care less about the framework name than whether you can see the whole path from input to consequence.

On the first call, teach us how you would debug an agent that retrieved the wrong account context and still sounded convincing. We should leave with a better test, not just a better explanation.