Presented practical patterns for building MCPs at duvo.ai, including in-house development for faster iteration, file system-based output storage, and using Skills to describe generated file structures.

Attended the Anthropic builder summit in London, exploring Anthropic's roadmap for AI agents and MCPs, and connecting with the LLM community.

Presented a poster at EuroPython benchmarking LLM API speed across 9 endpoints from OpenAI, Fireworks, and Anthropic over a full week, comparing tokens per second, response consistency, and answer length.

Built an AI agent at the E2B hackathon that automatically identifies owners of forgotten land plots by searching the web, indexing results, and cross-referencing databases. Won 2nd place.

This talk explored how LLM responses varied based on the query language, such as English vs. Czech. We analyzed the impact of language choice on model behavior and response costs.

I taught a group of 20 learners how to implement LLMs, covering agent programming, RAG systems, and embeddings.

At PyCon Estonia, I led a workshop on structured output and other new OpenAI features. I also gave a lightning talk on using Apple Silicon chips for LLM fine-tuning with MLX.

At EuroPython, I presented a poster describing how LLM performance differs when asked in various languages. I also looked at how recently released models tokenize text differently compared to their older versions.

This talk investigated how well Large Language Models work in various languages by discussing token usage differences, cross-lingual embedding effectiveness, and reasoning skills across languages. It reviewed models from OpenAI, Google, and open-source projects. I also led a RAG workshop focusing on the use of LlamaIndex.

The presentation explored LLM efficiency across languages, focusing on token economy variations, cross-lingual embeddings' effectiveness in bridging language gaps, and comparing reasoning abilities in different languages.

The presentation discussed the efficiency of GPT token encoding for non-English languages and introduced new OpenAI features like text-to-speech and Assistant APIs