Thank You Adi!

What We Learned While Developing with LLMs and NLP

At TEAMCAL AI, our journey to create ADI (Augmented Decision Intelligence) began with a simple yet ambitious question: How can we empower teams to schedule in seconds, freeing them to focus on their most important work?

Here’s a behind-the-scenes look at how we brought ADI to life and the valuable insights we gained along the way.

The Origins of ADI

When we first developed TEAMCAL AI, our scheduling platform, we achieved a breakthrough: reducing the average time it took to schedule a team meeting with more than 3 members from 7 minutes to just 2 minutes—a remarkable 72% reduction. This improvement saved teams countless hours every week. For example, our free Zoom App enables users to create follow-up meetings with multiple team members in just minutes while still on Zoom.

But then, AI happened. With advancements in natural language processing (NLP) and large language models (LLMs), we asked ourselves:

Can we leverage AI to reduce scheduling time to seconds?

This question led to the birth of ADI, driven by the BIG idea that advanced AI could achieve in under 30 seconds what previously took minutes. This vision became our North Star, guiding us to create a platform that doesn’t just schedule meetings but transforms how teams manage their time.

The early architecture of ADI SlackBot with the minimum feature set.

We Started with the Essential Features

• Text Command Support: Allows users to schedule meetings via typed commands.
• NLP and LLM Integration: Processes inputs, supports @names, and #team names, and resolves conflicts.
• Real-Time Availability Pooling: Aggregates participant availability across platforms to identify optimal meeting times.
• User Preferences: Considers constraints like preferred times and time zone compatibility.
• Platform Accessibility: Accessible through web, Slack, and Microsoft Teams integrations.

Then We Added the Magic

• Command Adaptability: ADI interprets ambiguous or incomplete user commands. For example, you can say, “Schedule a meeting for tomorrow,” and ADI will find a suitable time among all attendees and schedule it.
• Hypocorism Support: ADI understands nicknames commonly used in the workplace. For instance, you can say, “Schedule a meeting with Tom,” and ADI knows you mean Thomas. If there’s more than one Thomas, it will "find out" which one you mean.
• Fun Design: We incorporated a sleek and fun design for a more refined and enjoyable user experience!

What We Learned

• Focus on the Minimal Set of Features: By aligning every development milestone with a core user problem, we avoided feature bloat and ensured relevance.
• Eating Our Dog Food: Using ADI internally in our Slack workspace, we discovered its addictive potential. We also added a simple “Created by adi” tag to events, which helped us track its usage.
• Iterate Quickly: We adopted an iterative approach by quickly distributing the ADI beta to 10 of our existing customers. We then gathered their valuable feedback, which allowed us to make rapid improvements and fine-tune ADI based on real-world user experiences.
• Take one step at a time: The first version of ADI, a Slack integration, took six months to build. Slack Marketplace approval took another six weeks. During this short time, we were able to port ADI to Microsoft Teams also.

Unexpected Discoveries

One delightful discovery came from observing how users interacted with ADI. Many users would politely say:

“Thank you, ADI!”

However, the first version of ADI didn’t have a response for this! Seeing these interactions inspired us to incorporate more natural and conversational responses, making ADI feel even more human-like and approachable.

Now ADI will say something along the lines of

"You are welcome and good luck with your next meeting with Tom at 10"

Cool, huh!

The Voice and AI

The next big leap for ADI was the introduction of a Voice UI. When we got an opportunity to demo ADI at the Voice and AI Conference, we realized there was one problem: ADI was text-only and lacked a Voice UI.

We had one month to add voice capabilities. The challenge seemed insurmountable, with the team initially giving the project only a 10% chance of success within the tight deadline. My Job was on the line.

What followed was one of the most exciting and collaborative months we’ve ever had. We worked day and night, brainstorming use cases and refining interactions. Just three days before the conference, we did a final design and pulled an all-nighter to finish it.

And we did it. ADI’s Voice UI for Android debuted at the Voice and AI Conference, and the demo was pretty amazing. However, there were some challenges due to the limited range of the Wi-Fi. Oh Well!

How ADI Stands Out

ADI isn’t just a scheduling bot; it’s a holistic solution that combines NLP-powered conversational interfaces, real-time calendar analysis, and team-specific optimization. What makes it unique?

• ADI is Superfast: It cuts down the scheduling time by whooping 93% (30 seconds)
• Tailored for Teams: ADI understands team dynamics, priorities, and preferences.
• Smart Orchestration: It coordinates schedules across diverse time zones and organizational layers to maximize productivity.
• Secure by Design: Built with enterprise-grade security, ADI respects user data and integrates safely within corporate environments.

Looking Ahead: ADI 2.0

Agentic AI is the heart of ADI 2.0, the next step in our development.

ADI is going to transcend user input.

We aim to push the boundaries of what digital assistants can achieve. The uniqueness lies in how our system combines autonomous intelligence with user-centric design, ensuring reliability and trust while operating independently.

And Thank You, Dear Reader

If you made it this far, thank you. Help us by answering one of these questions

📌 What would you like to see in ADI 2.0?

📌 What additional features would make ADI even more valuable to your team?

📌 What impact AI-powered scheduling tools will have on the future of work?

Happy Friday!