Relari

🔍 How often do you RAG? No, it’s not something you’d use to wipe up a spill—it stands for Retrieval-Augmented Generation! Think of it as the AI's version of open-book exams: instead of generating answers from thin air, it “cheats” by pulling in relevant data from external sources. Even LLMs need some help sometimes... I'm excited to share the first episode of our limited series, GenAI Done Right! In this series, I sit down with fellow founders and execs to dive deep into the impact of Generative AI on business use cases, while tackling how to implement GenAI responsibly. In our debut episode, I had the pleasure of chatting with my friend, Yi Zhang, co-founder and CEO of Relari. Relari is revolutionizing how we ensure LLM-powered applications are safe and reliable, offering a data-driven toolkit to evaluate and improve them. Yi and I break down the ins and outs of Retrieval-Augmented Generation (RAG)—how it works, its pros and cons, and the critical criteria to consider when assessing GenAI products. I'll be posting more interviews with incredible AI entrepreneurs, so stay tuned for more insights on how to do GenAI right! 🎥 Catch the full episode here: https://lnkd.in/ecxSEfxC

I recently had a fun conversation with Darius from @TesoroAI where we reflected on some of the pivotal moments in my journey starting Relari with Pasquale. Some topics we covered: - My career trajectory before Relari - What motivated us to start a company in the AI infra space - What has been challenging about AI adoption despite all the buzz - Future of AI and Multimodal models If you're curious about the "why" behind Relari or want a peek into our journey, I’d love for you to check it out and share your thoughts! Youtube: https://lnkd.in/ekK2XAw5 Apple Podcast: https://lnkd.in/eubYnXCP Spotify: https://lnkd.in/e3EfbBxG

How do you measure the performance of a RAG app? We’ve been working with Relari to make that easier. In our latest blog, Thierry Damiba, David Myriel, and Yi Zhang show practical steps for: ✅ Adjusting 𝐓𝐨𝐩-𝐊 𝐩𝐚𝐫𝐚𝐦𝐞𝐭𝐞𝐫𝐬 to return the right number of results ✅ Using 𝐀𝐮𝐭𝐨 𝐏𝐫𝐨𝐦𝐩𝐭 𝐎𝐩𝐭𝐢𝐦𝐢𝐳𝐚𝐭𝐢𝐨𝐧 to refine chatbot responses ✅ Evaluating it all using the GitLab legal policies dataset as an example With this workflow, you can run fast, iterative tests, experiment with hybrid search, and access detailed metrics like precision, recall, and rank-based evaluations. Check out the full blog for code and implementation 👉 https://lnkd.in/dsw3BQrx

Few things make me happier than seeing our customers flourish and bring powerful AI applications to life.   Last week, Vanta raised $150 million at a $2.45 billion valuation, further solidifying its lead in the security compliance space.   Over the last several months, we've been working closely with Vanta's AI team. We’ve witnessed firsthand how effectively they leveraged a data-driven approach to productionize several high-impact LLM products. The result? Overwhelming positive feedback from Vanta's customers!   To date, Relari has proudly supported Vanta with over 50,000 evaluation runs on various synthetic datasets, ensuring robust performance for each LLM product.   A key recurring trend we've observed: contrary to popular belief, extensive testing accelerates iteration speed rather than slowing it down. Teams that implement data-driven tests from the prototyping phase reach production faster. They achieve this by making quick, confident decisions backed by rapid, targeted experiments on key components such as RAG parameters, prompts and LLM architectural choices.   Congratulations to Vanta on this milestone! We're excited to see the innovative Generative AI products they'll launch next.   Check out more details in this case study on our approach and results!   What trends have you noticed among teams building successful LLM products? What strategies seem to resonate most with customers? Share your insights in the comments! #LLM #RAG #GenerativeAI #LLMEvaluation #DataDrivenDevelopment

Another exciting opportunity to learn advanced RAG testing methods! Yi Zhang from Relari will be joining Thierry Damiba to discuss how synthetic data can be used to create production-ready RAG applications. This is a great opportunity for data scientists, ML engineers and developers working on GenAI to calibrate and improve their projects. The webinar will be followed up by a Q&A session, so bring your best questions and topics to discuss. See you on July 31st!