Unleashing Innovation: Insights from Adastra’s OpenAI Hackathon

August 17, 2023

OpenAI has emerged as a beacon of innovation, offering cutting-edge artificial intelligence solutions that have revolutionized the way we think, create, and operate. From the astonishing capabilities of language models to the advancements in generative AI, OpenAI has consistently redefined what is achievable in the realm of artificial intelligence.

To push the boundaries of generative AI and its ability to drive efficiencies for organizations, as well as foster a culture of creativity and collaboration, Adastra recently organized a company-wide Azure OpenAI hackathon. We saw some healthy competition as experts from our Data Science, Data Governance, Data Engineering, Azure, AWS, Application Development, GCP and Data Quality teams came together to develop their own solutions to solve business challenges using OpenAI.

In this blog post, we’ll dive into the exciting details of the event, highlighting the use cases that will have an impact on our customers and the valuable lessons learned.

Adastra’s Use Cases

Speeding Up Resource and Timeline Generation 

Organizations are busy, and project management is a function that can eat up large amounts of manual time and effort.

The first use case that emerged during Adastra’s hackathon is a tool to help organizations save time by more quickly generating project timelines and proposals. The use case allows users to input transcripts produced from client meetings held on Microsoft Teams to produce a project summary, required resources and the expected project duration, using OpenAI. It also enhances the quality of project proposals by combining the transcripts with previous notes and additional context.

The solution will save organizations from manually reviewing meeting notes, freeing up their time to focus on higher-value business activities.

Simplifying Content Creation

Content is an excellent way to deliver information, engage prospects and customers and build trust. But creating new content requires significant time and effort for research, creation and refinement.

The second solution that emerged used GPT as an accelerator to generate content based on input from question-and-answer scenarios and keywords, allowing for much quicker development.

Learn how Adastra can help your organization quickly produce targeted sales and marketing collateral that speaks to your customers.

Analyzing Customer Churn 

Customer churn is an important issue for any customer-facing industry.

Our third use case leverages OpenAI and organizational data to identify, with 90% accuracy, potential churners and recommend proactive retention strategies. The solution allows organizations to address customer needs and increase customer satisfaction, reduce churn rates, and reduce company retention and acquisition costs.

Standardizing ETL Processes

Developers often have many projects on the go, and performing manual extract, transform and load (ETL) processes slows down timelines.

The fourth use case from our hackathon standardizes and automates the ETL process based on source-to-target mapping (STM)’s. Our solution uses OpenAI to extract the latest data to be ingested from the source, generate transformation logic, and then upsert the data to the destination table, with 70-80% accuracy.

The solution can be used to speed up organizations’ path to data-driven insights by automating the process of preparing data for analytics including modeling, mapping and pipelining data. Learn more about how to stay competitive and unlock your data’s potential 10x faster.

Accelerating Application Testing

Many organizations develop applications internally, which need to go through user testing for quality assurance, but the testing process often requires significant manual labour with multiple stakeholders involved.

Adastra has developed a solution leveraging OpenAI to help with testing and internal documentation, resulting in saved time and effort for developers. Users can input their code directory and repository explanation, and OpenAI will work to detect any code changes and automatically generate user tests or code conversions as needed, ensuring consistent quality without manual time and effort.

Automating Data Cleansing

Having quality data (DQ) is crucial for organizations for many reasons, including producing accurate information for business decision-making, improving business performance, and ensuring regulatory compliance. But, a comprehensive data quality strategy can be costly.

Adastra’s sixth use case fast-tracks DQ rule creation in Java and profiling for data sets. This eliminates the need for any basic DQ functionality and reduces the time and money it takes for a DQ analyst or data engineer to analyze data and ensure it is quality.

Adastra’s Insights

Throughout the different hackathon solutions, a few common themes emerged.

Installing a Secure Instance of OpenAI

Many of the solutions involved entering confidential organizational data into OpenAI to produce the desired results. To ensure critical data is protected, the best approach is to install Azure OpenAI on your organization’s own private tenant, protecting it from potential data breaches and public consumption. Learn more.

Entering the Right Prompts

A common theme throughout the day was that prompts entered into GPT needed to be crafted and revised to produce the desired outputs. Sometimes this meant entering prompts in batches or providing additional information. For example, one team needed to devise a new 3-step prompt to give GPT the right context to produce accurate results.

Entering the right prompts when engaging with OpenAI is important for a successful and productive interaction, as the prompts serve as instructions to guide GPT’s response. If you are not getting the desired outcome from GPT, prompts may need to be re-engineered. Reach out to Adastra’s experts to work with us to craft the right prompts for OpenAI and begin unlocking the full potential of the platform.

Fact-Checking Outputs

Since GPT is trained on public data, it may pick up biases from this data, or may not be trained on the latest information. GPT itself asserts that it “may sometimes generate plausible-sounding but incorrect or biased information. Users should exercise critical thinking and fact-check information received from ChatGPT.”

In Adastra’s use cases, multiple teams spoke about the need to fact-check or have humans review GPT outputs. For example, during the hackathon, one of Adastra’s team compared outputs from OpenAI to previous reporting in Microsoft’s Power BI and found inconsistencies in its predictions.

To reduce the risk of false positives during code testing, the team provided additional context to GPT, combined with manual reviews of the generated tests.

Adastra also uses a Human in the Loop (HIL) framework to ensure accuracy, leveraging both machine and human intelligence to create, validate and update AI models.

Optimizing Costs

Using a private GPT instance is the safest and most secure way to leverage OpenAI within organizations, as when deployed on public ChatGPT, organizations’ sensitive data can be exposed to public consumption. However, private OpenAI costs money based on the size of the prompts, or the number of “tokens” entered.

During the hackathon, Adastra’s experts developed creative ways to optimize these costs. In the first use case, Adastra vectorized embeddings to reduce the character size of inputs into GPT and remove unnecessary information, or “fluff”. In the sample use case, the team was able to reduce their input into GPT from 6k to 2k characters, resulting in 74x cost savings.

To control spending, it’s important for organizations to work with an organization like Adastra to engineer prompts to produce the desired outcomes with the least number of tokens.

Discovering Untapped Opportunities for Automation

Adastra’s experts came up with numerous creative solutions, highlighting the many transformative opportunities and the potential of OpenAI to help businesses gain efficiency and save time in processes from all different organizational areas and functions. Despite the creativity we saw at Adastra’s hackathon, there are still many untapped opportunities to discover.

Not sure where to start? Work with Adastra to understand how your business can best benefit from OpenAI. Accelerate organizational processes safely and optimally with Adastra and Azure OpenAI.

Speak to our experts to get started.