First Engineering Night at Dust
We hosted our first Engineering Night in Paris on Feb. 12th bringing together over 80 AI engineers and researchers from across Europe for technical presentations and discussions. Your enthusiasm and feedback were incredibly motivating, and we're already planning the next edition.
Technical deep-dives with industry leaders from NVIDIA and Weaviate
The evening featured presentations from three industry leaders who shared their expertise and insights.
Jonathan Raiman from NVIDIA discussed "RL at Scale," exploring how to scale reinforcement learning systems through advanced reward modeling and policy optimization techniques. His presentation highlighted how Group Relative Policy Optimization (GRPO) improves training efficiency by reusing successful trajectories and leveraging dense rewards through process supervision. The talk demonstrated how modern RL systems can achieve better performance by combining automated verification for ground truth with reinforcement learning, enabling more efficient exploration of solution spaces and better generalization beyond training examples.
Daniel Madalitso Phiri from Weaviate introduced "Agentic RAG," exploring recent advancements and methodologies in retrieval-augmented generation. He began his presentation by discussing traditional RAG systems and their growing importance in modern AI applications. The focus then shifted to Agentic RAG, an agent-driven version of RAG that incorporates agents into the pipeline. This innovative approach enhances capabilities beyond simple information retrieval and effectively addresses the limitations of non-agentic systems.
Our very own Alban Dumouilla demonstrated "Query Tables" which introduced a system that transforms natural language into SQL queries across diverse data sources (warehouses, spreadsheets, CSVs). The technology unifies different data formats, generates appropriate SQL code, and creates visualizations automatically. This approach enables seamless data analysis without coding knowledge, effectively delivering "natural language business intelligence" capabilities to non-technical users.
The presentations led to a technical Q&A session with our co-founder Stanislas Polu and the guest speakers, with many questions ranging from RAG implementation challenges, RL applications, infrastructure choices around vector databases and LLM operations to why we chose to open-source our code.
"Being open source is not a strategy but rather a proactive posture towards openness. Openness for our users and our ecosystem (sharing PRs when fixing bugs, pointing to our codebase in our communication) and openness for our security enabling ethical researchers to push our infrastructure to its limits." Stanislas Polu, Co-founder, Dust.
Casual conversations, networking, drinks and pizzas
What would be a good ‘night’ if not followed by drinks and pizza? We’ve loved watching everyone continue the discussions for hours. Some went deep into technical details; others wandered about AI and its overall impact.
Looking ahead
Again, a huge thank you to all our speakers and attendees who made this evening special. Many of you expressed appreciation for the event which convinced us to plan future editions. If you attended, we'd love to hear your feedback to improve future events in this short survey. Thanks to those who’ve already responded!
Join our community and stay connected
- Track upcoming Dust events on our Lu.ma page
- Follow us on LinkedIn and X
- Explore opportunities to join Dust
We're looking forward to the next edition!