From Hack Days to Innovation Labs: Putting AI Models to the Test

What’s it like to work at Oomph? It’s about more than just delivering world-class digital experiences. It’s about being part of a team that’s given the freedom to explore, experiment, and innovate. Enter: Innovation Labs — our twist on the classic Hack Day, designed for a remote-first workplace.

The Spirit of Innovation

At Oomph, we know great work starts with great opportunities. That’s why we give our team dedicated time to develop ideas, play with new tools, and push their skills in new directions. Our Innovation Labs are quarterly events where the entire company collaborates on solving real-world challenges — while having a lot of fun in the process.

For our first Innovation Lab this past fall, the mission was simple: use generative AI (GenAI) to save at least one hour of work per week. With AI reshaping the future of web development, this challenge gave everyone — from designers to project managers — the chance to learn about cutting-edge tech and apply it in meaningful ways.

How We Innovate

Here’s how we structured the Innovation Lab:

• Six Teams: We mixed project managers, designers, developers, and sales team members into diverse teams.
• Six AI Models: Each team was assigned an AI model, from OpenAI to Llama, to tackle a specific problem.
• One Day: Starting at 10:00 a.m. ET, teams brainstormed, built, and tested their ideas. By 4:30 p.m., it was time to present their work.
• Support & Fun: We kept the creativity flowing with Giftogram cards for meals and a celebratory toast at the end of the day.

What We Built

The projects were as diverse as the AI models we worked with, showcasing the creativity and problem-solving power of the Oomph team:

1. Team OpenAI: Project Estimating

The team developed EstimAItor, a bot designed to make project estimating more efficient by analyzing historical task data from Jira. They exported a year’s worth of task titles, descriptions, and time log data, integrated it with OpenAI’s API, and built a Python-based app. The bot could estimate task durations, provide confidence levels, and identify related issues. The result? A tool that eliminates guesswork and streamlines the estimation process.

2. Team Anthropic Claude: Project Estimating

Similar to Team OpenAI, this group leveraged Anthropic’s Claude model to connect with Jira and estimate task durations. Claude was tasked with analyzing past data to validate and refine project estimates. While the tool wasn’t fully functional by the end of the day, the team identified several exciting use cases, such as using Claude to double-check estimates against past projects, combining data fields (like titles, descriptions, and time logs), and producing insightful summaries.

3. Team Hugging Face: Meal Prep

This team aimed to tackle a time-consuming part of everyday life: meal prep. They used the Hugging Face Inference API along with meta-llama and Llama-3.2-1B-Instruct models to scrape data from recipe websites like AllRecipes. Their tool could simplify recipes, generate concise shopping lists, and help users reduce time spent on meal planning. While the AI successfully summarized recipes and provided shopping lists, the team noted some uncertainty about whether the data extraction was based on the recipe URL or similar recipes in its training set.

4. Team Mistral: Project Estimating

Using Mistral, Jira, Postman, and GitHub, this team created a tool to make task estimation more precise and context-specific. Their solution used historical data to suggest time estimates, which users could either accept or refine further. The team also explored how the tool could help assign the right team members to specific tasks based on past work. By eliminating manual guesswork, this tool could save significant time and effort for project managers.

5. Team Gemini: Meal Prep

Gemini’s project added a dose of personality to meal planning. Using a photo of the user’s fridge contents, their tool suggested recipes — complete with a cheeky Gordon Ramsay-inspired twist. For example, if the fridge contained eggs, milk, and vegetables, Gemini might suggest an omelet while delivering a signature Ramsay quip. The tool assumed basic pantry staples (like flour and rice) and allowed users to customize meal types and dietary preferences. When one of the team members took a photo of their fridge, it even recognized cat food in the photo. Thankfully, Gordon didn’t suggest a cat food souffle. 

6. Team Llama AI: Time Tracking

The Llama Log tool automated time tracking by connecting GitHub commits with Jira Tempo timesheets. Using GitHub Webhooks for notifications and the Jira Tempo API for updates, the team streamlined the process of logging hours for developers. Despite running into storage limitations with Ollama and compatibility challenges with Meta models, the team successfully demonstrated how AI could automate workflows and even integrated branding and content generation into their project presentation.

The Winner: Team Gemini

At the end of the day, the entire Oomph team came together to present their projects and vote for the winner. No judges, no top-down decisions — just a democratic process where every voice counted. The winning project? Team Gemini, whose creative and functional recipe tool stood out for its innovation and personality.

Ready to Innovate?