ExperimentingwithGenerativeAI:Dynamo6’shackathon
Ever wondered how a company tackles new and emerging technologies? At Dynamo6, we embrace innovation. That's why we recently held an internal hackathon focused on Generative AI, a powerful new tool with the potential to revolutionise various industries.
This blog post dives into our journey – from setting the challenge to the winning solution – showcasing the exciting possibilities Generative AI holds for the future.
Skip ahead
What we came up with:
Painting the scene
With our company origins based in digital transformation via cloud technology, the latest innovation to hit the world Generative AI presents a turning point. Harnessing new technology, making bold changes, and disrupting the status quo are all options on the table.
But Generative AI is a new technology. It presents challenges of understanding exactly what it is, how it can be harnessed, and the real business value it can offer. That’s what we wanted to experiment with.
So, to do that, we held an internal hackathon.1 A dedicated 8-hour competition, to pick up a real-world business challenge, think about how technology could solve it, create a Proof of Concept2 in various stages, and finally pitch the idea Dragon’s Den style to a panel of judges.
Setting the challenge
Drawing inspiration from our recent client projects as well as our internal support desk systems, there were four available challenges for the teams to choose from.3
Before the hackathon day, all participants were kept in the dark about who their teammates were going to be and what the challenges were. Behind the scenes, the format, available challenges and prizes were set.
We shared a few guides about what a hackathon is, captured the team’s ideas for challenges they’d like to tackle, as well as a handy guide from Google Cloud on GenAI.4
Our friends and technology partners at Google Cloud also supported us on the day. Providing us with guidance on the application of Vertex AI to our chosen challenges, as well as judging the four team’s end-of-day pitches.
Lifting the cloak
We kicked off the day with breakfast and a briefing. Across 8 hours, we’d form into four multi-disciplinary teams, pick a challenge to tackle and use the day to explore and come up with a solution and work to create our Proof of Concepts.
And here’s what we came up with…
OIA Smart Search
A smart assistant to improve the process and turnaround time for Official Information Act (OIA) requests for government organisations. The OIA Smart Search assistant would simplify and improve the responding body’s sourcing of information, being quicker and helping to deliver a more streamlined and overall better user experience to the requester.
Examples of it in action:
- Summarise meeting content topics and minutes across various committees.
Provide summaries of topics that have been discussed in meetings or documents when asked. For example, ‘What’s the latest traffic management update for Pukete?’
- Search across previous requests, identifying information that’s been provided before or is similar, to provide it readily again to the requester.
- Draw from internal datasets to provide answers with minimal internal processing. For example, “How many parking tickets have been issued this year?” And provided drafted answers ready for an internal compliance or response team to review before being issued.
- And if a query hasn’t been discussed or asked before, or otherwise can’t auto-generate a response, provide an intuitive guided form, allowing the requester to submit a new request to follow the standard 20-day OIA request process.
Support Goose
Support Goose is a smart assistant ready to deliver a first response to a customer query submitted to an IT helpdesk system.
Unlike present-day chatbots, Support Goose is an internal assistant. It acts as a co-pilot, offering helpful support to the Support Agent responding to customer queries. The Proof of Concept build drew on Dynamo6’s internal service desk systems and knowledge bases, and could readily add publicly available information to help quickly respond to IT support issues.
The co-pilot drafts appropriate, human-like responses, fed with intelligence from the internal and external data, which the Support Agent validates before sending to the customer.
Percentage changes and actual responses are further added to the system to further improve the model. Additional response feedback is given by the Support Agent; Good Goose (thumbs up) and Silly Goose (thumbs down), naturally.
Kiwi AI
A smart AI assistant named “Kiwi AI” that helps people looking to buy shoes across a range of shoe stores in shopping centres.
Drawing on the Cloud Vision technology, taking a photo of a shoe you like and sharing it with the Kiwi AI assistant, any available shoe styles, prices, retailers and availability will be presented, with the standard handy size, colour preference and relevance filters we’re used to using when shopping online.
Here are some specific tasks that the assistant could help with:
- Finding shoes by brand, style, or size
- Compare prices between different retailers
- Read reviews of shoes
- Find out if a particular shoe is in stock, and on sale
- Help with placing an order for shoes, which could be online or in-store
The assistant could deliver an immersive experience for shoppers across shopping centres, making it easier for them to find the perfect pair of shoes across the range of stores available to them to see and try on in-store or order online.
Legal AId
The law is complex. Knowing how to comply with local bylaws that impact building works and improvements can be a costly and confusing process for the average citizen.
- Can I build a sleepout on my section?
- Do I need planning consent to install an ensuite?
- Or how far does my chuck house need to be away from the boundary fence?
But imagine a legal aide, ready to answer your questions. A smart assistant able to draw from local bylaws and match them with regional and national rules and regulations, and provide non-legalise guidance if you can do that home improvement you’ve been thinking about or not. All without needing to wade through document after document, translate jargon, or spending an arm and a leg on legal or planning fees.
The winner
Based on strict judging criteria, including evaluating the technical details, the market opportunity, and the overall pitch given by the team that worked on the Proof of Concept, the judges voted OIA Smart Search as the winner.
It was incredibly close, and ultimately, the real triumph lay in the collective efforts of all involved, working together to create working technology concepts and getting to grips with something completely new to technology, Generative AI.
Footnotes
1 A hackathon is an event designed to use technology to accomplish an objective. It’s a cross-functional event, where teams of people and skills such as software engineering, product management, design, and many more are all working together and supporting the creation of functioning software or hardware by the end of the event.
2 A proof of concept can be defined in a number of stages, for example:
- An explained idea, technical summary, and feasible plan.
- A sample of the idea, in code or concept drawings, designs, or descriptions.
- A coded element of the solution.
3 These were the four challenges the teams could pick:
- A smartbot that could answer any homeware appliance questions or troubleshoot problems.
- An assistant that could search for information on a city council’s website providing answers to a ratepayer’s questions or interpreting data sets.
- A smart assistant that could look across a mall's website, provide answers to any questions and helping the customer find products they’re after in stores they can visit.
- And finally, option 4 was a ‘first responder’ to any IT problems. A smart system that looked across internal system knowledge and publicly available information to provide quick responses to any technical challenges.
4 Google Cloud, ‘The Executive’s Guide to Generative AI‘, published 2023.