|
Marija Maneva
Data Science · AI Engineer
Hi, I am Marija. I have a master's degree in Data Science (Computer Science) and I've worked both in a startup and a consulting company. My goal is to create something valuable in the tech world.
|
|
|
MSc. Computer Engineering (Data Science)
University of Pavia · Italy
Thesis: "A Deep Learning Framework for Vector Graphics Generation using Stable Diffusion"
BSc. Bioengineering
University of Pavia · Italy
Thesis: "Machine Learning Techniques for the Prediction of the Progression of Multiple Sclerosis"
|
|
Master Thesis — Stable Diffusion for Vector Graphics
Developed a full AI generation system from text prompts with complete data pipeline. Fine-tuned Stable Diffusion on a custom dataset, deployed inference endpoint, and implemented AI-assisted prompt refinement with a user-friendly interface.
Movie Recommendation App
Conversational recommendation application using generative AI for personalized movie suggestions through natural language. Showcases RAG architecture and multi-source data integration.
Large-Scale Movie Reviews Analytics
Conducted large-scale data analysis on movie reviews using big data technologies. Demonstrated distributed computing and cross-platform processing.
Cloud Computing — Scalable Distributed App
Developed a distributed, scalable, elastic, and fault-tolerant application leveraging AWS cloud services. Production-grade microservices with serverless architecture and infrastructure automation.
|
Logo AI Generator
An AI tool that generates logo designs by fine-tuning a deep learning model on a dataset of existing logos. It was my first hands-on project with generative image models, built during my time at Produkto. I prototyped it with Gradio and trained it on Google Colab, using pretrained models from Hugging Face as a starting point.
Skill Agent
A cloud-native, multi-tenant conversational profiling system built to support SDLC adoption across different clients, now used by thousands of people. I worked on this end to end, from the database schema and migrations all the way to the Angular front end. I also set up the CI/CD pipelines and monitoring that keep it running in production.
Proposal Validation Agent
A multi-agent pipeline that automates the validation of business proposals, with specialised agents for rule extraction, tagging, conflict detection and verification. It took a process that used to take days and brought it down to hours for a team of business analysts. I instrumented the whole pipeline with Langfuse and RAGAS, so quality is tracked automatically rather than checked by hand.
Conversational and Estimation Agents
A business analyst chatbot built around a dual-memory system, combining short-term conversation context with longer-term structured memory stored in a graph. This lets the agent hold a natural conversation while still handling estimation tasks that need consistent, structured knowledge. Built with LangChain, Neo4j, FastAPI and Python.
Document Intelligence: File Parser and Processor
A hybrid document extraction pipeline that handles PDF, DOCX and PPTX files by first analysing how complex each page is. Simple pages go through a lightweight text model, while more complex ones get routed to a vision model. That routing logic alone cut inference costs by close to half without losing accuracy.