Hi,
Over the last three months, I’ve been working on Agents and LLM workflows. I had the opportunity to do this as part of a residency program at Merantix. I’m deeply thankful for the opportunity and would highly recommend the program to others. Being surrounded by people with shared goals was incredibly motivating and inspiring. I already miss the discussions and the knowledge exchange. Unlike typical work environments, here the participants were truly aligned, and everyone learned from each other.
My result of the last three months: link
I worked on:
- implementing a tool-based agent
- experimenting with LLM workflows
- adding an LLM-as-a-judge evaluation framework
- building an event-driven architecture around the agent and workflows
To support this, I created a custom environment with APIs, RAG,UI and synthetic data stored in a database.
My learnings:
- Agents will be the next step of automation - the challenge is setting them up efficiently
- Evaluation will be the backbone of effective agents - getting this right is hard but essential
- Claude Code is awesome - both a blessing and a curse
- Building something people actually want is hard - technology alone isn’t enough
What I can improve:
- Spend less time on glue code - if the product has no value, nothing else matters
- Invest more time in networking - conversations help to uncover real needs
- Focus on core technical challenges - this is where learning and value starts for me.
- Sharpen on product thinking - do I just like the idea or does it really solve a problem?
Next steps:
- Dedicate more time on ideation
- Explore LLM-as-a-judge with open-weight models for less biased, probability-based evaluation
- Keep solving real problems - learning happens by doing
Thank you for your attention.