I’m Djenis Ejupi, a software engineer in Switzerland. I work across backend systems, applied ML, automation and product design. This is where I share the projects, questions and experiments that are mine.
Making local LLM analysis useful, inspectable and dependable as a complete desktop utility.
01 / BUILDTurning daily life in Switzerland into deliberate language practice rather than another passive goal.
02 / LEARNMoving towards systems that sense, decide and act in the physical world, with clear limits and human control.
03 / EXPLOREI’m maintaining six open-source projects, deepening my work in backend engineering, cloud, DevOps and ML systems, and learning German.
My longer-term interests include robotics, control systems, autonomous systems and human-centred technology.
Longer case studies live in the shared Ejupi Labs journal, filtered to personal projects.
Each project has working software, inspectable source and release evidence. The point is the decision the software improves, or the boundary it makes visible.
A local-first desktop career utility. Analysis, matching and coaching require an approved local LLM runtime, with no cloud fallback.
A Rust open-set ML workbench with group-aware splits, calibration, abstention and byte-verifiable artifacts.
Windows and browser automation where the model proposes an action and the runtime enforces permissions, capabilities and allowlists.
A bounded Gopher client and protocol explorer. The CLI uses real connections; the browser demo uses deterministic fixtures.
A Java and TypeScript workbench for comparing numerical integration methods against a visible Simpson reference.
A browser-only placement operations board with fictional data, explicit cohort states and local persistence.
Clear boundaries, useful tests, safe failure modes and interfaces that explain themselves matter as much as the visible result. I choose the tool after I understand the constraint.
Software is the medium I use today. My longer-term interests reach into robotics, control systems, intelligent automation and human-centred technology.
Useful first. Evidence over claims.
A system should say what it can do, what it cannot do and where it needs a human decision.
Tests earn their place when they protect a decision, an interface or a failure mode that matters.
I prefer explicit refusal, bounded work and recoverable state over a result that merely looks confident.
The interface should make the model of the system easier to understand, not turn it into decoration.
I studied Computer Science and Telecommunications, then worked across cloud infrastructure, enterprise software, backend and frontend delivery, and workflow platforms.
A technical education grounded in software, networks and how systems connect.
Database, backend, frontend and workflow delivery inside systems used every day.
Repeatable infrastructure, delivery pipelines and workflows that make operational state visible.
Building tools where model limits, local execution and evidence are part of the product.
The common thread is end-to-end ownership: understand the constraint, change the system and prove it still works.
This inbox is for conversations with me. For client work, proposals or delivery enquiries, use Ejupi Labs.