Warsaw, Poland · Machine learning since 2017
AI Research Engineer & Technical Lead
I design and build evaluation infrastructure, agentic applications and ML systems, from Responsible AI evaluation at eBay through Tooploox to technical leadership in biopharmaceutical manufacturing at Roche.
My independent research examines small language models in multi-turn workflows: preserving information, producing correct structured outputs and balancing task quality with latency, memory and energy.
Selected research & engineering contributions
01I built a local information-collection prototype and an evaluation framework to study multi-turn state, structured outputs and inference trade-offs. Later experiments examine routing decisions and instruction boundaries.
Read the research →As a Senior AI Research Engineer through Tooploox, I translated research methods into RAMP components: endpoint evaluation, model judges, automated red-teaming, benchmark coverage and review dashboards. The work brought evaluation evidence into operational AI risk review.
See my eBay contribution →I led ML/data architecture across manufacturing sites in a regulated GxP environment, supported models of titer and yield, and led engineering delivery. I also built a GenAI/RAG starter platform and authored a playbook for architecture and evaluation.
Explore Roche experience →I help energy-tech and automotive startups design and build secure AI systems. My work spans local and cloud inference, retrieval, voice interfaces and agent workflows, with structured outputs, explicit conversation state, observability and application-specific evaluation.
Explore startup consulting →I work across experiment design, evaluation methodology, software architecture and implementation. My background combines empirical ML, production engineering and leadership in regulated environments.
Professional experience
Evaluation infrastructure, biomedical ML systems and technical leadership across industry settings.
2026 - present · Warsaw, Poland
Senior AI Research Engineer
Research engineering and technical leadership for RAMP, connecting Responsible AI evaluation, benchmark coverage and evidence-based review within eBay’s AI Platform.
Explore my work
2021–2024 · Basel, Switzerland
Global Senior Data Engineer · Data Mesh Technical Delivery Lead
Federated data and ML architecture across manufacturing sites, GenAI foundations and leadership of engineering delivery in a regulated GxP environment.
Explore my work
December 2024–present
AI Research Engineer Consultant
Research and consulting across local models, agentic systems, retrieval and voice applications for energy-tech and automotive startups.
Explore my work
2017–2020 · Warsaw
Data Scientist
Predictive modelling, statistical comparison and data engineering, turning financial data into evidence for business decisions.
Explore my work
August–September 2020 · Project
Data Scientist · Computer Vision
Deep-learning-based detection of power-pole faults from drone imagery, from raw images to automated identification.
Explore my work
May 2016–September 2017
Talent Development Program
Statistical investigation of manufacturing parameters, component consistency and line performance.
Explore my workIndependent research
Small language models, multi-turn information collection and the systems that make their behaviour measurable.
Complete research article
A comprehensive study of small language models for multi-turn information collection: evaluation methodology, model comparisons, inference trade-offs, later experiments and lessons learned, with the results and figures in one article.
Read the full story →Independent engineering
I build and use agentic applications across career development, research, content, learning, finances and everyday planning.
My projects combine conversational interfaces with persistent memory, retrieval, domain-specific tools and reviewable outputs. Building them for recurring tasks gives me a practical way to examine how agents handle information, retain context and carry work across sessions.
CareerApp · Web & messaging
An agentic career application with long-term memory: structured career stories, hybrid retrieval, targeted interviews and reviewable CV generation across the web, Telegram and WhatsApp. Plugin development is the next extension.
Read the full story →Content experimentation
An end-to-end content experimentation toolkit: episode scripts, ElevenLabs audio prototypes, account analytics, hypothesis tracking and cross-platform video production, connected in one feedback loop.
Read the full story →Research & knowledge
A daily research agent that checks fresh sources, connects developments to my projects and delivers a concise written and spoken briefing.
Read the full story →Learning & feedback
An interactive learning application that executes Python, visualizes recursion and backtracking, and compares a handwritten trace with the actual program.
Read the full story →Money & administration
A tool-backed workflow for budgets, invoices, payments and documents, with a structured ledger, reviewable imports and editable spreadsheet outputs.
Read the full story →Learning & personal memory
A conversational learning workflow that turns real writing corrections into persistent lessons, recurring language patterns and spoken revision.
Read the full story →Search & everyday decisions
Browser-assisted research for comparing travel options with real dates, prices and exact offer links, bringing structured evidence into a planning conversation.
Read the full story →Education
Select a university to see my field of study, specialization and completion status.
Links & resources
Resources grouped by topic, with a description of what each link contains.
Methods, findings and the repositories behind the work.
Application architecture, multi-turn evaluation, model comparisons, runtime experiments and later routing studies.
The complete research article, figures, configurations and retrospective.
Python evaluation framework, experiment definitions, response captures and scoring.
Flutter prototype with local inference and structured conversation state. GitHub access may be required.
Tools and applications I build.
Long-term memory, career-story retrieval, web, Telegram and WhatsApp workflows, voice and plugin development.
Use the conversational career assistant on careerapp.io.
Episode scripts, ElevenLabs prototypes, account analytics, hypothesis tracking, editing and cross-platform republishing.
A sourced daily AI briefing with project-specific context, spoken delivery and persistent delivery state.
Interactive Python execution, recursion traces, backtracking lessons and model-assisted feedback.
Budgets, invoices, payments and reviewable document imports connected to a structured ledger.
Persistent lessons, recurring language patterns and spoken revision from real conversations.
Browser-assisted offer discovery with dates, prices and exact destination links.
Contact
Get in touch about research, AI engineering or building a system together.
maria.jastrzebska.ai@gmail.com
For research, engineering and consulting enquiries.
Connect on LinkedIn
Professional experience, recommendations and public activity.
Explore my engineering work
Software projects, research code and technical documentation.
@Marijas_ai
Find me on X.
