> projects

Things I've built.

Production systems, research projects, and things that started as experiments and became real.

> featured

Womli

launching april 2026solo project

AI companion app built specifically for women. Conversational, contextually aware, built on a RAG pipeline that actually remembers what matters. This is the app I wanted to exist. I built it myself — product, backend, mobile, infra, everything.

stack
React NativeExpoFastAPISupabasepgvectorClaude Haiku 4.5sentence-transformersRAG PipelineRevenueCat
key decisions
Claude Haiku over GPT
Faster response times at lower cost for conversational flows. GPT-4 latency was too noticeable for a companion context.
pgvector inside Supabase
Keeps the vector store co-located with user data. Fewer moving parts. Supabase RLS handles per-user access control automatically.
Local sentence-transformers
No API calls for embedding generation means lower latency and cost. The model runs server-side — user data doesn't leave my infra.
GitHub — private during developmentApp Store — coming April 2026
> other work
CoreMagnet / VU Amsterdam

Bachelor Thesis — MCP + Claude AI Workflow Automation

Researched and implemented Model Context Protocol integration with Claude to automate enterprise workflows. Evaluated accuracy, latency, and agent reliability across real business processes.

MCPClaude AIAutomationLLMPython
view →
Floorplanner

RAG Pipeline — Design Dataset Search

Built a production RAG pipeline for searching a large design asset dataset. Implemented hierarchical chunking strategy to preserve document structure and improve retrieval relevance.

RAGPythonHierarchical ChunkingEmbeddingsVector Search
VU Amsterdam

Predictive Models — Video Game Dataset

Applied ML classification and regression models to predict game sales and ratings. Compared model performance across feature engineering strategies.

Pythonscikit-learnPyTorchMLData Analysis
view →
VU Amsterdam

NLP / Text Mining Project

Built a text mining pipeline including tokenization, POS tagging, NER, and sentiment classification. Evaluated multiple NLP architectures on a domain-specific corpus.

NLPspaCyTransformersPythonText Mining
view →
VU Amsterdam

AI Law & Ethics

Research and analysis on legal and ethical frameworks for AI systems.

AI EthicsResearchPolicy
view →
VU Amsterdam

Algorithms & Deep Learning

Deep learning implementations and algorithm analysis.

Deep LearningPyTorchAlgorithms
view →
VU Amsterdam

Statistics & R

Statistical analysis and modeling in R.

RStatisticsData Analysis
view →
VU Amsterdam

Collective Intelligence

Research into collective intelligence systems and emergent behavior.

ResearchMulti-agentCollective AI
view →