Conference Presentation, Keynote
'Production Ready RAG' by David Leconte from DataStax | RAISE Summit 2024 | Paris
- Datastack, with over a decade of enterprise machine learning experience, is pivoting to address the complexities of deploying production-ready RAGs, noting that 14 years of Apache Cassandra knowledge underpins its infrastructure.
- The company forecasts that organizations will face significant challenges in identifying root causes and preventing failures in broken applications due to the asynchronous development and frequent API changes of hundreds of open-source libraries.
- To mitigate risks associated with unsupported open-source components, particularly in regulated banking markets requiring validation, Datastack offers RAGStack, a solution bundling curated libraries from frameworks like LangChain and LlamaIndex.
- RAGStack undergoes daily testing and thousands of security scans to prevent SQL injection and cyber risks, with a commitment to collaboratively resolve issues with framework editors when conflicts arise.
- The platform enables a single
pip install -r ragstackcommand to de-risk predictions, eliminates the need for manual library hunting, and ensures continuous functionality even with external API failures from providers like OpenAI. - Datastack provides a 24-7 "follow-the-sun" support model and an on-premise release designed for high throughput, low latency (targeting 10 milliseconds), and scale.
- The company offers a serverless, multicloud, and multi-region database service that leverages 10 years of enterprise data experience to avoid bottlenecks in embedding, ingestion, and retrieval.
- Independent studies cited claim Datastack is 74% faster than dedicated vector databases while maintaining superior accuracy and F1 scores compared to competitors like Pinecone.
- Longflow, recently acquired by Datastack, provides a workflow engine and UI to generate well-written code and facilitate deployment via API, aiming to reduce development time in Gen-AI driven decision cycles.
- Strategic positioning includes integration with NVIDIA's NEMO framework to accelerate embedding mechanisms at lower costs, while also offering alternatives to closed stacks for companies in regulated environments.
- The outlook warns that companies failing to adopt streamlined solutions face risks of reinventing the wheel or engaging in difficult forking processes, which could result in losing the race for innovation.