stativa(7) · ai/ml engineering · bucharest
I work across AI/ML and backend systems — data pipelines, ML models, and reliable APIs. This site collects a few notes, project results, and practical experiments around retrieval, inference cost, software systems, and measurement.
second pass — zero-framework lite shell
A small map of writing about AI systems, retrieval, measurement, infrastructure, and the costs behind GenAI work.
Hover a node to isolate its thread; click to read. † = forthcoming.
A simple model for comparing GenAI request strategies by business value, inference cost, and risk. Drag the slider to see how the values change.
EV = p·bv − cost − risk risk_reserve = $0.05 p: A 0.00 · B 0.67 · C 0.67
At $0.42 business value, retrieval flips EV positive; fine-tuning breaks even after 106 requests.
| request type | small model | RAG + mid | large reasoning |
|---|---|---|---|
| password reset | +$0.031 | +$0.025 | −$0.018 |
| policy question | +$0.012 | +$0.146 | +$0.091 |
| account exception | −$0.041 | +$0.064 | +$0.173 |
A practical view of the languages, infrastructure, search tools, and ML libraries I use or study regularly.
Python daily · ai/ml 
PHP production · 10y 
TypeScript learning · ui 
Scala production · spark 
Golang learning · backend 
Docker daily · infra 
Apache Solr production · search 
Elasticsearch production · search 
Spark/ML production · ml 
PyTorch learning · models 
CUDA learning · systems 
LangChain learning · agents Most work starts with a clear problem, a small prototype, or an existing system that needs careful improvement.
Predictive models, data pipelines, inference economics, and production ML systems for real workloads.
REST APIs, microservices, batch processing, and reliability work for products that need to keep moving.
Architecture review, TDD practices, observability, and delivery support for teams improving reliability.
Performance, monitoring, search systems, and measurable improvements for production platforms.
A compact overview of recent roles, project work, and education.
AI Engineer, CedaraMoved an advertising optimization agent from PoC toward production and connected ML decisioning to programmatic bidding workflows.
Maintained carbon-monitoring analytics pipelines and expanded social-media training datasets. Python, PyTorch, Scala, SparkML, Dataproc, OpenX API, Laravel, Airflow, GCP.
Senior Backend Engineer, PentalogConsulted for a major European investing platform inside a service-oriented architecture.
Maintained 99% unit test coverage, supported bi-weekly releases with 0% error rate, and added Datadog observability. PHP, Laravel, MySQL, Docker, Redis, Kafka.
Senior Backend Engineer, IsoSkills / EzugiRe-architected reporting reads onto a replica to reduce primary database load and improve operational headroom.
PHP, Symfony, API Platform, MySQL, Docker.
Delivered an admin platform for gas and energy providers with tariff logic and operational workflows.
React, Redux, PHP, Symfony, API Platform, MySQL/MariaDB, Docker.
Backend Engineer, Bandai Namco Entertainment RomaniaBuilt a RESTful API from scratch for Retro Wonder Park, supporting backend gameplay and mobile integration requirements.
Senior Backend Engineer, eMAGHelped extract the search API from a monolith into a dedicated Apache Solr microservice.
Won the Future Hacks 2.0 Innovation Award for funnel-stage prediction and reduced regression runtime by 90%. R, Python, Docker, SeleniumGrid.
Moved from critical fixes into core gaming-platform development, player tracking, and QA automation training. Java, Selenium, Git.
Delivered outsourcing projects including Rolii.ro, a legal portal indexing 5M+ court decisions, and Multibonus.ro, a MOL loyalty platform.
Early roles at WebVenture, Customsoft, and The Shop Page. PHP, Agile.
Send a short note about the project, the problem, or the kind of help you need.
Built with Astro, self-hosted fonts, subset icons, and hand-authored CSS. This page ships through a gated preview route before any production port.
