About Me
I am an AI researcher who also builds the production systems that make research usable outside a paper. The doctorate gives the credential; the discipline is already running in public, one pre-registered experiment at a time.
Research
From October 2026 I am a doctoral researcher at the Institute of Robotics, Bulgarian Academy of Sciences, in the Interactive Robotics and Control Systems section. My topic is adaptive large language models for social robotics through continual learning and knowledge retention mechanisms.
In plain terms: a model can learn something new, but in doing so it tends to overwrite what it already knew. For a chatbot that is an inconvenience. For a robot working with one specific person, adapting to their pace and their good and bad days, it is a dealbreaker. The same question runs through my engineering work: how do you build systems that keep learning without becoming unpredictable.
Before the doctorate starts, the method already runs in public. Every experiment on this site is pre-registered, with a kill-gate threshold written down before any code runs, and published with the same weight whether the gate passes or fires. Frozen plans and full result tables are in a public repository; the write-ups are under Research on this site.
Engineering
For more than 20 years I have been writing production software: PHP and Laravel services, Python APIs, integrations, and the unglamorous infrastructure that keeps them running. Since 2024 I have been at euShipments, a cross-border e-commerce fulfilment provider, where I started on the distributed logistics platform and its integrations with more than sixty courier companies. Over the past months my work there has shifted almost entirely to AI: designing and running the integrations and automations across the company's processes, from support workflows to document reconciliation and internal data access.
That is also where most of my current interest sits, where AI meets real operations. In practice it means retrieval-augmented systems grounded in a company's own documents rather than in a model's memory, processing pipelines that survive messy real-world formats, anomaly detection over structured data, and masking layers that keep sensitive information from ever reaching an external model provider.
Since 2018 I have also worked as a contractor through my own company, AFC EOOD, founded in 2004. This is the work that funds the research and keeps the engineering honest: systems built for real operations, not demos.
Background
I hold an MSc in Artificial Intelligence and Data Processing and a BSc in Information Technologies, both from the University of Library Studies and Information Technologies in Sofia.
Selected work through AFC EOOD:
- Acumen Risk, a risk management platform built with More By Us, where field reports from industrial sites are turned into models and signals that help managers decide which protective measures to strengthen. Its users include teams at BP and OMV. Angular and Neo4j, later moving to React and MongoDB
- UR Fog, a manufacturer of active protection systems in Turin, Italy. I built the cloud portal for one of their IoT boards, collecting data from and sending commands to customer machines connected to their alarm systems
- Luximmo and Suprimmo, a content management system used to build and launch a dedicated site for each new building project sold off-plan
Earlier work as an employee:
- Dir.bg, one of the most visited news portals in Bulgaria. I redesigned their sites and moved them onto a new editorial platform used daily by journalists and editors
- GreenDC, an EU-funded project analysing and forecasting electricity consumption in the data centre of the Turkish telecom operator Turksat, delivered as part of the DAVID Holding team
- Admissions and student records portal for the Higher School of Telecommunications and Posts, and other public-sector systems
This site
I write here about pre-registered experiments and what they actually found, including the ones that did not work, alongside notes on applied AI in production and backend architecture. If you are a research collaborator, or building AI inside your organisation, the fastest way to start is an email.