S
SINAPTIA
Guest
On August 28th, 2025, we sponsored and attended the 3rd AI4Devs meetup of the year. It was an amazing event, fully packed:
The meetup kicked off with Gustavo Archuby and Pablo Musa from the Facultad de Humanidades, UNLP. They talked about their experience building a complete RAG system for an academic document base for research and preservation. They built it exclusively with open source tools and open weight models and everything self-hosted. They also talked about the challenges of data consistency, data fixing for effective indexing and retrieval.
The second presentation was Ulises Cornejoβs. He talked us about building rulesets for code assistants, focused more than anything in the interactions and overlap of docs for AI tools and for humans. He higlighted several conflicts,and pain points, and proposed a system to have a single source of truth that works for both human developers and AI assistants, that leave no one out of the picture.
The last talk came from our very own Fernando Martinez. He talked about pains and lessons learned from SINAPTIAβs successful experiences building AI-based features. He touched topics like testing non-deterministic systems, evals and test dataset building; challenges and solutions needed when using the OpenAIβs batch API to process millions of requests per month, and a couple of fun anecdotes about pricing and the ultimate need for monitoring.
As always there was time to have a nice beer and sandwiches and catch up with other folks in the industry. There were a lot of people we havenβt seen in years! And the variety of people (professors, students, profesional) was a fantastic mix. Props to the organizers from AI4devs La Plata
was a fantastic night.
We look forward anxiously for the next one!
Continue reading...
- almost one hundred atendees ranging from young students, university professors, and developers of every seniority
- 3 interesting talks and speakers
- and, as the icing on the cake, hosted on a venue called βSobre Rielesβ (Coincidence? I donβt think so!)
RAG applications on local infrastructure
The meetup kicked off with Gustavo Archuby and Pablo Musa from the Facultad de Humanidades, UNLP. They talked about their experience building a complete RAG system for an academic document base for research and preservation. They built it exclusively with open source tools and open weight models and everything self-hosted. They also talked about the challenges of data consistency, data fixing for effective indexing and retrieval.
Itβs not magic, itβs engineering
The second presentation was Ulises Cornejoβs. He talked us about building rulesets for code assistants, focused more than anything in the interactions and overlap of docs for AI tools and for humans. He higlighted several conflicts,and pain points, and proposed a system to have a single source of truth that works for both human developers and AI assistants, that leave no one out of the picture.
LLM Workflows for classification and large-scale data coherence
The last talk came from our very own Fernando Martinez. He talked about pains and lessons learned from SINAPTIAβs successful experiences building AI-based features. He touched topics like testing non-deterministic systems, evals and test dataset building; challenges and solutions needed when using the OpenAIβs batch API to process millions of requests per month, and a couple of fun anecdotes about pricing and the ultimate need for monitoring.
Community and Networking
As always there was time to have a nice beer and sandwiches and catch up with other folks in the industry. There were a lot of people we havenβt seen in years! And the variety of people (professors, students, profesional) was a fantastic mix. Props to the organizers from AI4devs La Plata
was a fantastic night.We look forward anxiously for the next one!
Continue reading...
