by Ed Lenta, June Tan, Jacintha Ng and Brian Law
We're excited to announce the winners of Databricks' inaugural Asia-Pacific Large Language Model (LLM) Cup - in partnership with AWS, a first-of-its-kind competition in the region, which garnered participation from over 1,000 data and AI practitioners across more than 10 countries.
From October to December 2023, participants were invited to build LLM-powered applications using Databricks to solve real-world business problems. To help set up participants for success, prior to the hackathon, we provided self-paced learning workshops on building LLMs and dedicated training led by Databricks' solution architects.
Entries were evaluated based on creativity, business applicability, relevance, thoroughness and the scalability of the LLM architecture. While there were many of impactful projects, two teams stood out with their innovative ideas addressing critical organisational challenges in telecommunications and cybersecurity.
Congratulations to the winning teams whose innovations prove just how quickly teams can transform businesses with the right technology!
Securing the grand prize, Contact Catalyst from Optus, an Australian telecommunications company, addressed a prevailing business challenge: understanding customers' needs during calls. Call centre agents typically take manual notes, leading to unreliable interpretations, often resulting in customer frustration and dissatisfaction.
Using synthetic data sets, the team analysed data in an automated and secure fashion, and navigated the complexities of managing vast volumes of text and voice data from customer interactions and gained deep, real-time insights into customers' needs, significantly enhancing the overall user experience.
Its LLM can efficiently classify the intent of the customer call, provide real-time detection of new topics, content summarisation sentiment analysis and detect aggression, alerting the business to address the customer's concerns quickly and politely.
The runner-up, Red Teaming Your LLM by Ensign InfoSecurity ("Ensign"), Asia's largest pure-play end-to-end cybersecurity service provider, demonstrated their solution to counter the cyber vulnerabilities with growing deployment of LLM-based applications.
With greater accessibility to LLM APIs, more companies are building their own LLM-based applications. Yet, the rapid deployment of these models has outpaced the development of comprehensive security protocols, leaving many applications vulnerable to high-risks cyberattacks. The Ensign team devised a "red teaming approach" to train LLMs with proprietary jailbreak prompts. Consequently, the LLMs proactively identified and addressed vulnerabilities in LLM defences, enhancing overall cybersecurity.
We are thrilled to see so much interest and innovation coming from the data and AI community across APJ. It's inspiring to see teams use Databricks to develop LLMs that tackle their unique challenges – really getting to the heart of what they need to better serve their customers and revolutionise their business operations.
Both teams presented their winning entries at Databricks' APJ Data + AI World Tour on Thursday, 18 January 2024 (recording available on-demand).
Visit the project gallery to learn more about their submissions: https://databricks-llm-cup-2023.devpost.com/project-gallery.
We'd also like to express our gratitude to AWS for their partnership in the LLM Cup - playing a crucial role in the event’s success.
The Databricks Asia-Pacific LLM Cup was the first competition of its kind in the region, run in partnership with AWS from October to December 2023. It drew participation from over 1,000 data and AI practitioners across more than 10 countries, who were invited to build LLM-powered applications using Databricks to solve real-world business problems. Ahead of the hackathon, participants had access to self-paced learning workshops and dedicated training led by Databricks solutions architects.
Entries were evaluated on creativity, business applicability, relevance, thoroughness and the scalability of the LLM architecture. Judges considered how well each solution addressed a genuine organisational challenge alongside the technical soundness of its design. Two teams stood out for tackling critical challenges in telecommunications and cybersecurity.
Optus, an Australian telecommunications company, won the grand prize with Contact Catalyst, an LLM-based solution that gives a 360-degree view of call centre customer conversations. Using synthetic data sets, the team analysed vast volumes of text and voice data in an automated and secure fashion to classify customer call intent, detect new topics in real time, summarise content, analyse sentiment and flag aggression so the business can respond quickly and appropriately. This replaces the manual note-taking agents typically rely on, which often led to unreliable interpretations and customer dissatisfaction.
Ensign InfoSecurity, described as Asia's largest pure-play end-to-end cybersecurity service provider, took runner-up with Red Teaming Your LLM, a solution addressing cyber vulnerabilities created by the rapid, security-outpacing deployment of LLM-based applications. The team used a red teaming approach, training LLMs with proprietary jailbreak prompts so the models could proactively identify and address vulnerabilities in LLM defences.
Both winning teams presented their entries at Databricks' APJ Data + AI World Tour on 18 January 2024, and a recording is available on-demand. You can also view full submission details for Contact Catalyst and Red Teaming Your LLM, or browse all entries in the LLM Cup project gallery.
Subscribe to our blog and get the latest posts delivered to your inbox.