The situation
Dimitar led work on an AI-enabled platform for understanding legacy software. The challenge connected software analysis, enterprise context and a usable architecture for real delivery.
A proof of concept built independently
The first idea was a chat interface over an existing internal tool. Dimitar built the first proof of concept independently, analysing codebases directly. It surfaced hidden business flows and system capabilities, not just file summaries.
From chat wrapper to platform
The evidence from the proof of concept moved the initiative from a chat wrapper to a code-intelligence platform. Once the business agreed, Dimitar designed the target architecture and led the work to production, with a team scaled to 16+ AI engineers.
The architecture contribution
The work brought together multi-agent analysis, retrieval and graph-based structures to investigate software. Architecture decisions had to address integration, evidence organisation and the practical operation of a production platform.
Into enterprise conversations and pilots
The platform was taken into many enterprise conversations and commercial pilots. That gave a direct feedback loop between product design, sales conversations and real transformation use cases.
What the pilots showed
Delivery teams started with a fuller picture of unfamiliar systems. The pilots supported the central idea: AI is most valuable when it helps architects understand the real system beneath the documentation.
Why this experience matters
It informs Sparkling's focus on connecting code-level evidence with system understanding and an accountable architectural interpretation. Tools support that work. The conclusions still depend on an architect checking them in context.
The architecture lesson
Software intelligence is useful when it improves a real architecture or modernisation decision.
Led by Dimitar Stoyanov before Sparkling was founded.
