Dashworks, your AI-powered internal search tool 🔍

Plus: Founder Prasad on the opportunity around internal search...

September 04, 2024

CV Deep Dive

Today, we’re talking with Prasad Kawthekar, Co-Founder and CEO of Dashworks.

Dashworks is an AI-powered internal search and answer assistant designed to help teams find and access information across their various tools and systems. It aims to address the challenges of scattered and siloed data within organizations, integrating with a wide range of tools, from documentation and wikis like Notion and Google Drive to project management tools like Jira and CRM systems like Salesforce, providing real-time, synthesized answers to internal questions.

Dashworks is used by companies of all sizes, from fast-growing startups like HeyGen and Synthesia to Fortune 500 enterprises, enabling teams to quickly access the information they need, whether it’s for sales, support, or engineering. The platform’s unique approach to real-time API-based search, which avoids the need for pre-indexing, sets it apart in terms of security, data freshness, ease of setup, and cost. Dashworks is focused on building what they call the “truth engine” for enterprises—a reliable source of information that will eventually power automated workflows and actions across different systems.

In this conversation, Prasad takes us through the founding story of Dashworks, the challenges of building a real-time internal search platform, and what’s next on the roadmap for the company.

Let’s dive in ⚡️

Read time: 8 mins

Our Chat with Prasad 💬

Prasad - welcome to Cerebral Valley! First off, give us a bit about your background and what led you to found Dashworks? 

My background is in AI research and engineering. I was one of the founding NLP engineers at Cresta, where I got to see firsthand what LLMs could do for enterprise productivity and collaboration. At Cresta, we were building language models for sales—like, creating responses to sales questions. These weren't the large language models we know today; they were much smaller. The "Attention Is All You Need" paper had just come out, so there was a lot of buzz about the potential of the Transformer architecture. We started applying language models to some use cases with enterprise customers, and the results were pretty impressive. That’s how I first got into NLP.

Before Cresta, I was already working on AI and machine learning, doing deep learning at Stanford, and before that, at Carnegie Mellon. I met my co-founder, Praty during Stanford grad school, where he was also working on machine learning and deep learning projects. Eventually, we both decided to move into the industry.

During this time, I noticed that it was easy to find information on the web using a search engine, but within work applications, it was a nightmare because everything was scattered across different systems. I remember in one of my roles, it took me seven weeks to find one document because it was hidden in some Box folder. That experience really highlighted a problem I wanted to solve.

As I started digging into this problem more, I realized that with everything happening in NLP, we could finally tackle some of the big challenges in enterprise search. So, I reached out to Praty—who was working on search at Facebook at the time—and we decided to team up and take this idea to YC.

Give us a top level overview of Dashworks - how would you describe the startup to those who are maybe less familiar with you? 

Dashworks is an AI that helps teams get answers to their internal questions. It can search through all the internal tools a team uses—whether it’s documentation in Notion, Google Drive, Dropbox, Box, internal chat systems like Slack and Microsoft Teams, project management tools like Jira and Linear, CRM systems like Salesforce, or even code bases like GitHub. It pulls information from all these different systems and synthesizes it to provide a clear answer to whatever internal question you have.

Who are your users today? Who’s finding the most value in what you’re building with Dashworks? 

We work with a variety of companies, from fast-growing startups like HeyGen and Synthesia to medium and large enterprises. Our use cases vary—support and sales reps often need quick answers to product or technical questions, engineers might be trying to resolve bugs or issues, and new hires benefit from a centralized place to find information during onboarding. Information is usually scattered across different tools, so having a single place to look things up is crucial. We often work with client-facing teams, like sales, support, and enablement, who need quick answers to internal questions, especially when they're dealing with customer issues or technical questions that need to be resolved quickly.

There are a number of teams working on knowledge workers across various enterprise domains. What sets Dashworks apart from others in the space? 

One of the unique aspects of Dashworks is that we’ve optimized it so it doesn’t require indexing your internal company data. A common concern with solutions like ours is around security and privacy. To address this, we’re SOC 2 Type 2 compliant, HIPAA compliant, and GDPR compliant, and we have all the appropriate DPAs in place. But what really sets us apart is that Dashworks functions as a real-time search agent. It looks up information live, rather than pre-indexing all of your data into a single service.

This approach not only enhances security and compliance but also makes Dashworks faster to set up—teams can be up and running in just a few minutes since there’s no indexing time required. Plus, because it performs live searches, you’re always getting the latest information, not relying on a stale index. It’s also more cost-effective, as we pass the savings from not having to index data back to our customers.

Our product is hyper-focused on a specific use case today: internal question answering. We think of ourselves as building a "truth engine" for enterprises. Our ultimate goal is to enable the execution of actions and workflows across different systems. For an agent to reliably run actions on behalf of a customer—whether it’s deploying to infrastructure or communicating with customers—it needs to know the truth within the company. It needs to know the correct answer to questions about products, projects, processes, or policies.

Internal documentation can often be messy, with outdated, missing, or conflicting information. A lot of our work goes into resolving these discrepancies and figuring out the right ranking and retrieval across different systems, whether it’s a Jira ticket, a Slack message, or a Confluence wiki. The idea is that by doing this, we build an agent that customers can trust to take reliable actions on their behalf.

How do you measure the impact that Dashworks is having on your key customers? Any customer success stories that you’d like to share?  

One example I’d highlight is our work with Podium, one of the largest and fastest-growing YC companies with about 1300 employees. They found themselves managing around 20 to 25 Slack channels where people were constantly asking questions about different product features, integrations, support-related issues, and sales-related matters. These channels had names like “ask-feature-X,” (for their different product lines) and employees were frequently asking product managers, engineers, support leads, and enablement leads for help.

As you can imagine, this was a huge time sink—not just for the reps who were waiting around for answers but also for the people responding to the questions, who spent a lot of time addressing repetitive queries. After implementing our Slack bot, Podium saw some pretty dramatic results. They were able to automate about 87.8% of the questions being asked in these Slack channels. They also cut down the response time and the wait time for answers by about 96.2%.

We’ve heard some interesting stories as well. One of the product managers at Podium, before going on paternity leave, actually included in his leave doc that Dashworks would be handling all the questions in his absence for the next couple of months. It really shows how our solution automates away a lot of the internal busywork around answering the same questions repeatedly, and it ensures you always get quick, canonical responses by searching all the relevant documentation.

There’s been an explosion of interest in agentic workflows. How has that shaped the way you’re thinking about building Dashworks? 

We don't have agents built into the system just yet, but Dashworks is designed to directly interact with the APIs of different apps that companies use, like Slack, GitHub, Google Drive, and Jira. Dashworks makes real-time API calls to pull together the right answers to various questions, so in that sense, it's agentic. It doesn’t rely on a static index stored within the system. However, we’re not at the point of taking actions or running workflows just yet, but that’s definitely something we’re working towards.

One of the main reasons we chose this architectural approach—using real-time API calls—was to pave the way for a future where we can do more than just read from APIs. We want to be able to write back and potentially take multiple steps if necessary. Expanding from just read APIs to write APIs and from single-hop to multi-hop processes are areas we're currently focusing on. We believe that building this foundation now, what we’re calling the “truth engine,” is crucial for creating reliable agents that enterprises can trust to get work done. If an agent doesn’t fully understand your codebase, processes, or customers, for example, in the future you wouldn’t trust it to deploy code to production to fix a customer issue. So, the agent needs to have a solid grasp of what's true and accurate within an organization to be trustworthy in making decisions and taking actions.

What has been the hardest technical challenge you’ve faced so far while building Dashworks? 

Since this is a new space, there are a lot of interesting challenges to tackle. I’ll highlight a few of them. Our real-time API approach requires us to rethink how to optimize latency and accuracy. We've done a lot of novel work around this, and we think our systems are state-of-the-art when it comes to accuracy and handling rate limits. We're constantly working on improving latency as well.

Because we rely heavily on Retrieval-Augmented Generation (RAG) in our systems, we're always looking to enhance the instruction-following capabilities of these systems and their ability to understand large context lengths. We’re also exploring fine-tuning open-source systems like the new Llama 405B model to see how it can further improve our performance in these areas.

Another major challenge in enterprise settings is the lack of good, clean documentation. There’s often a lack of incentive to create and maintain documentation, so we have to rely heavily on ranking and retrieval methods that can deal with these issues by using various signals—like who created the documentation, when it was created, where it’s located, and so on. We spend a lot of time fine-tuning our retrieval ranking engines to handle these challenges effectively.

How do you plan on Dashworks progressing over the next 6-12 months? Anything specific on your roadmap that new or existing customers should be excited for? 

Our customers are really excited about the direction we're heading. One key area is moving from single-hop read systems, which is what Dashworks primarily does today, to more multi-hop capabilities. This shift would allow us to tackle more complex questions, perform deeper analyses, and synthesize data across various tools and databases. It’s about addressing more sophisticated queries.

The other big focus is automation. Right now, Dashworks reads information and presents it, but the next step is enabling it to take actions on your behalf. For example, not just writing an email draft for you, but actually sending that email. Or not just identify a solution to a customer issue but also directly respond to the customer and resolve the issue. We're really looking forward to advancing in these areas, enabling more complex question handling and action-oriented automation.

You went through YC in 2019. Tell us a little bit about that experience - how instrumental was YC in helping you shape Dashworks’ early direction? 

YC was incredibly helpful for us. Their emphasis on shipping early and often, getting the product into the hands of users quickly, and maintaining a growth-focused mentality was crucial. Without that push, we might have spent way more time just building in isolation without getting real feedback from users and customers. That focus on rapid iteration and user feedback was probably the most valuable thing we took away from YC.

Lastly, tell us a little bit about the team and culture at Dashworks. How big is the company now, and what do you look for in prospective team members that are joining?

We're a small but driven team based in San Francisco, right near South Park. Right now, we're actively hiring for three roles: a software engineer with a focus on machine learning, a full-stack software engineer, and a product designer. We like to have the team come into the office, and we're definitely on the lookout for talented individuals to fill these positions.

We're especially looking for self-starters who are genuinely excited about solving customer problems and can move quickly. We love working with people who can make sound decisions and execute independently because the space we’re in is evolving fast. We're really optimizing for iteration velocity, so anyone who thrives in that kind of environment is likely to be a great fit.

Conclusion

To stay up to date on the latest with Dashworks, learn more about them here.  

Read our past few Deep Dives below:

If you would like us to ‘Deep Dive’ a founder, team or product launch, please reply to this email ([email protected]) or DM us on Twitter or LinkedIn.