DevCrawl Session #3, track 2 - Optimizing RAG Workflows

It's Notch DevCrawl time again! This meetup, hosted by our AI experts Darko Špoljarić and Marco Hrlić, explores cutting-edge techniques for building and optimizing Retrieval-Augmented Generation (RAG) systems.

Across both sessions, you’ll gain insights into modern approaches that make AI pipelines more reliable, scalable, and efficient—from smarter evaluation methods to new frameworks that bring structure and clarity to working with large language models.

Once the talks wrap up, stick around—this is where the magic happens. We’ll have space for questions, casual chats, and those spontaneous “what if we tried…” conversations. Call it networking, call it mingling… we just call it the best part of the night.

NOTE: Here's a simple RSVP link for those who prefer not to deal with required subscriptions, CC details & co.

AGENDA

18:00 - 18:45 Generating synthetic test data for your RAG pipeline: DeepEval vs. Ragas, with Darko Špoljarić.
The manual generation of training and test data for AI-powered systems is a time-consuming and tedious task. There are various concerns to address, including labeling, completeness, filtering, bias, and quantity. However, when implementing a RAG system, there's already a dataset considered to be of "golden" quality. It's used to generate a RAG context from, so why not use that corpus to create the test data from? Come and learn more about the approach and how the two most popular frameworks today compare in that particular area.

18:45 - 19:15 - Refill and munchies break!

19:15 - 20:00
Declarative Self-improving Python (DSPy), with Marco Hrlić.
DSPy addresses the challenge of constructing compound AI systems by employing a declarative approach, as opposed to 'fuzzy' programming through natural language prompts. Interactions with LLMs, including instructions, context, data structures, and other aspects, are explicitly defined in the code. We'll cover the core principles and design philosophy of the framework by designing and building a simple system to solve a common RAG task. We'll conclude with an overview and a demonstration of optimization techniques offered by DSPy.

20:00 - 21:00 - Mix & Mingle! Grab a beer and let's talk some more.

SPEAKERS:

Darko Špoljarić, Vice President of AI
Darko has 15 years of experience in enterprise software development with Java. He has worked as a software engineer and team leader at top Croatian tech companies and is a certified Agile coach, one of only 15 Certified Scrum Professionals in Croatia. His passion for machine learning started early; he took Andrew Ng’s course back in 2016, and this year, he hit a significant milestone by becoming Notch’s first Vice President of AI.

Marco Hrlić, AI Engineer
Marco started his career at Ericsson Nikola Tesla as a student, developing embedded software for radio units. At Spotsie and Notch, he helped design and implement an IoT Bluetooth Low Energy positioning system. His skills include product and hardware design, embedded software development, cloud data processing, network protocols, AI engineering, mathematical modeling, and algorithm development.

Find out more about:

DevCrawl
From the idea to a complete solution, from tech rookie to AI master, and from first meeting to deployment, we encounter many starts and finishes during the software development process and our careers. We like to "cheer" each time we cross one. And, to be fair, is there a better way to discuss any topic than over a beer?

Notch
Notch is an AI-first enterprise software development and consulting company dedicated to creating robust, innovative, and secure digital solutions tailored for global enterprises. Recognized as a leading software developer and AI-strategy firm in several international rankings and proud to be a Deloitte Impact Star laureate, Notch combines decades of collective experience, ensuring long-term success for its clients.

The event will be recorded.
By attending our DewCrawl event, you agree to be photographed and recorded on audio-video for promotional purposes. The promotional videos and photographs may be published and reproduced in various media, such as social networks, the Internet, and the press.