DBot / Alpega

Applied AI in transport operations

At DBot Software, I worked as a Senior Full Stack / AI Engineer on Alpega's transport platform from April 2023 to October 2025. My contribution connected Java development, enterprise integration and context-aware guidance for drivers, consultants and transport teams.

Problem and operating context

The nearest service station is not always the right one. Decisions depend on carrier position, shipment state, operational history, station capability, timing and availability. Discovery should lead to an operationally useful recommendation and a clear booking outcome.

My responsibility

I contributed backend services, REST integrations and discovery, guidance and booking functionality. I designed a four-milestone Hybrid Search/RAG workflow and connected it to shipment visibility and ranked recommendations. My delivery scope also included automated QA, regression prevention, CI/CD and stabilization.

System architecture

Shipment context and operational history inform retrieval. Station capabilities and process requirements support matching and ranking. The application connects guidance to existing business services and booking behavior. This is a responsibility view, not a disclosure of internal endpoint names or deployment topology.

Technical decisions

I combined complementary context instead of reducing the problem to a location match. Current state establishes the immediate need; history adds context; station capabilities determine fit. Established Java/Spring Boot services keep the intelligent workflow connected to the surrounding enterprise application.

Implementation details

The workflow covered loading, unloading, reloading and cleaning recommendations. Discovery incorporated more than 20,000 cleaning-station locations, including information relevant to EFTCO, EFTCI and ENFIT processes. Location, time, availability and process fit were ranking considerations.

Production challenges

An API response and a completed booking describe different outcomes. Integration behavior must communicate what actually happened. Defect investigation crossed application and service boundaries, while controlled releases and regression coverage supported ongoing operation.

Evaluation and observability

My work included automated QA, API and integration tests, regression runs, technical documentation and CI/CD. The supplied manuscript describes an internally measured workflow improvement; it does not provide a public experimental dataset or a reproducible model benchmark.

Verified metrics and attribution

20,000+ locations describes the workflow scope. Four milestones describes the Hybrid Search/RAG design. The supported workflow recorded an internally measured improvement above 8% in transport time and operating cost. This is not a platform-wide outcome.

DBot separately publishes 42% higher performance, 99.95% uptime and 30% lower infrastructure cost for the wider engineering engagement, which involved more than 15 full-time developers. These belong to the team engagement, not to my individual contribution. The supplied manuscript cites DBot's case study and engagement results; its source notes retain the measurement limitations.

Technology stack

Java, Spring Boot, REST services, enterprise integrations, Hybrid Search, RAG, automated API/integration testing and CI/CD. My work combined domain understanding with full-stack delivery and operational support.

Diagrams and evidence

The original project illustration and responsibility overview come from the supplied ten-page portfolio. The full manuscript distinguishes personal engineering responsibility, internal workflow results and published team-level outcomes.

Lessons and next steps

Useful applied AI begins with a business decision and continues through the software that executes it. The engineering lesson is to keep context, recommendation, business action and result connected. Future evaluation should preserve those distinctions and establish a documented baseline before expanding any performance claim.