About

David Papp

AI engineering student at VU Amsterdam, building production-quality tools for the LLM stack. I focus on the infrastructure between models and users — observability, guard rails, fine-tuning pipelines, and retrieval systems.

Background

I consulted for an AI-first startup where I restructured their LLM backend to cut unnecessary OpenAI API calls by ~40% and closed a prompt injection vulnerability in their moderation layer. I consistently take the technical lead role in hackathon and rapid-build teams of 4–5 people.

Before university, I worked as a Project Assistant at 4iG (Budapest) on two large public-sector IT projects, and taught programming at Logiscool.

Education

BSc Artificial Intelligence — Vrije Universiteit Amsterdam

Sep 2023 – Expected 2026

Machine Learning · Deep Learning · NLP · Computer Vision · Data Mining · Algorithms & Complexity · Statistics

Advanced Studies in Computer Science — Milestone Institute, Budapest

2019–2023

Led the Computer Sciences Society.

Skills

AI & LLM Tooling

Daily use

Prompt engineering, structured outputs, tool calling, guard rail design.

See: MCP Sentinel, RAG Chat

OpenAI APIAnthropic APILangChainCrewAIHuggingFaceStructured Outputs

RAG & Retrieval

Active

Embedding pipelines, vector search (pgvector), cross-encoder reranking, citation enforcement.

See: RAG Chat

pgvectorSupabaseOpenAI EmbeddingsHybrid Search

Agentic Systems & Security

Daily use

MCP protocol, injection detection, PII scanning, moderation hardening, cost-based rate limiting.

See: MCP Sentinel, Startup consulting

MCPGuard RailsHMAC SigningEvent Logging

Backend & Data

Active

Full-stack TypeScript/Python. Auth, database, rate limiting, and deployment.

See: All projects

PythonDjangoTypeScriptNext.jsPostgreSQLSupabaseAWS LambdaDynamoDBDockerKafka (basic)

Frontend & Infra

Working

Built this portfolio platform end-to-end: auth (Clerk), database (Supabase + RLS), rate limiting (Upstash), and deployment (Vercel).

See: This platform

ReactTailwindVercelCloudflareClerk

Model Training

Active

Built the dataset-to-LoRA pipeline: codebase ingestion, AST-aware chunking, JSONL formatting, and fine-tune job management.

See: Training Pipeline

LoRA / PEFTJSONLAST ChunkingHuggingFace Hub

Availability

Based in Rotterdam, Netherlands. Available for part-time roles during studies, full-time from mid-2026. Open to: junior AI/automation engineer, full-stack AI, and data science roles.

About — David Papp