Collato: Enabling effective knowledge management with Vertex AI
About Collato
Founded in 2020, Berlin-based startup Collato aims to make the future of work more human-centric. The company's AI-enabled software revolutionizes the way product teams create documentation, ensuring essential knowledge is captured and accessible to all team members.
Tell us your challenge. We're here to help.
Contact usCollato migrated its infrastructure to Google Cloud to access advanced fine-tuning capabilities and the latest LLMs, while protecting user data with robust encryption.
Google Cloud results
- Empowered Collato's ML team to undertake complex fine-tuning tasks
- Enabled an effective data distillation strategy
- Protects user data with robust, transparent encryption standards
Fast-tracked Collato's ML success by six months
Knowledge is power, and all the more so in the workplace. Research shows that knowledge sharing increases creativity, fosters innovation, and boosts performance across the board. However, effective knowledge management can come with significant challenges, especially in an age of ever-increasing data volumes, multiple communication channels, and remote or hybrid working.
The Berlin-based startup Collato was founded in 2020 by Sebastian Oehlschläger and Ivo Scherkamp with the aim of redefining how teams work together. They soon realized that the best way to do this was to improve access to information across organizations. "During the pandemic, collaboration tools that had existed for years really took off," says Oehlschläger, Co-founder and CTO, Collato. "But more tools also created more information silos. Collato enables teams to instantly capture knowledge, using AI to distill essential information into cohesive, user-friendly formats."
Collato wanted to enhance its solutions with cutting-edge ML technologies, so the company migrated its infrastructure to Google Cloud.
Staying on top of a fast-evolving field with Google Cloud generative AI solutions
"We focus on knowledge creation for teams, especially product managers, who conduct a lot of research and are in constant conversation with users and other stakeholders," Oehlschläger explains. "The result is lots of scattered information in different formats, whether it's sketches, notes, or meeting transcripts. Collato uses AI to transform that information into accessible knowledge." Collato's AI assistant allows users to instantly transform raw information (including audio recordings, text, and images) into structured, polished documents. Customizable templates enable users to create requirements documents, one-pagers, user research summaries, product guidelines, and support documentation in seconds.
"Imagine you're working at a large retail company and you're developing a VR feature for virtual dressing rooms," Oehlschläger says. "You would need to take into account existing knowledge about warehouse operations, order placements, and payment processing as well as your own notes. In order to brief designers and engineers, that information needs to be condensed into clear and comprehensive documents." This process is at the heart of Collato's mission. The company transforms scattered information into accessible knowledge that saves workers significant time in the document creation process.
In 2023, Collato began a six-month journey in which it migrated its entire infrastructure to Google Cloud, making Vertex AI the backbone of its solutions. "AI is a fast-moving field, and not only do we need to be fast, everything we work with does as well. We need the latest language models, and access to GPUs. Being held back in one area can set us back significantly," Oehlschläger says. "Initially, we were keen to work with Google Cloud because its vector-based solutions fit our needs. But what made a real difference for us was early access to the latest language models for text generation and embeddings."
Delivering advanced fine-tuning capabilities with VertexAI
The Google Cloud account team supported the startup to become part of the Trusted Tester Program, which gives it access to the latest Google Cloud generative AI solutions. Collato's AI assistant is built on the Vertex AI platform, using Vector Search to store and retrieve knowledge artifacts and PaLM 2 and Gemini to understand user questions and generate answers. Before migrating to Google Cloud, the company had worked with another AI company. "With VertexAI, there was a significant improvement. The models were far easier to fine-tune. We are able to use smaller models, enhance their performance and tailor them to specific tasks. We now get comparable results with much shorter response times," says Oehlschläger.
"With VertexAI, there was a significant improvement. The models were far easier to fine-tune. We are able to use smaller models, enhance their performance, and tailor them to specific tasks, and get comparable results with much shorter response times."
—Sebastian Oehlschläger, Co-founder and CTO, CollatoCollato uses a data distillation strategy, in which large models generate training data that is then used on smaller models, which in turn operate much faster. "Fine-tuning large language models was something we had planned for later on in the process, possibly once we'd expanded the team. Once we saw how much easier things were with Google Cloud, we decided to just go for it," says Oehlschläger. The ability to move quickly and efficiently with limited resources is a core strength of startups like Collato, allowing them to find their niche in the market. The Collato team believes that working with Google Cloud accelerated their machine learning journey by around six months and brought their new product features to the market faster.
Protecting user information is another major priority for Collato. "Security is extremely important to us, as our customers trust us with access to their work tools. Google Cloud offers a secure setup with transparent encryption standards, which is reassuring for both us and our users," says Sara Kola, Senior Software Engineer at Collato. The company was also reassured by Google Cloud's commitment to not use user-provided data for training purposes.
"Fine-tuning large language models was something we had planned for later on in the process, possibly once we'd expanded the team. Once we saw how much easier things were with Google Cloud, we decided to just go for it."
—Sebastian Oehlschläger, Co-founder and CTO, CollatoLooking forward to advanced vision recognition capabilities with Gemini
Since the migration, Collato has seen considerable improvements in latency, fewer hallucinations, and improved LLM accuracy. The company is continuing to fine-tune models to tailor them to specific tasks. Short-term plans include exploring Gemini's advanced vision recognition capabilities, empowering its AI assistant to interpret more subtle visual inputs, such as the direction of arrows in a sketch or how post-its are clustered on a vision board.
Work with open source models provided by Model Garden is also in the program. "So far, we've focused on fine-tuning Google's proprietary models, but expanding to open source models is a big next step," Oehlschläger explains.
The Collato team worked closely with Google Cloud throughout the migration and were impressed with the level of support they received. "We're a small team, and working with Google Cloud has made it possible to achieve so much more than we initially thought possible," says Ivo Scherkamp, Co-Founder and CEO, Collato. "We're now able to deliver smart, capable tools that make a real difference to our users."
"We're a small team, and working with Google Cloud has made it possible to achieve so much more that we initially thought possible. We're now able to deliver smart, capable tools that make a real difference to our users."
—Ivo Scherkamp, Co-Founder and CEO, CollatoTell us your challenge. We're here to help.
Contact us
About Collato
Founded in 2020, Berlin-based startup Collato aims to make the future of work more human-centric. The company's AI-enabled software revolutionizes the way product teams create documentation, ensuring essential knowledge is captured and accessible to all team members.