DutyDingo, a startup founded by two seasoned software development agency owners, approached us with a vision torevolutionize project management in the tech industry. Their goal was to streamline the often cumbersome process ofcreating tickets from chat conversations and managing project information more efficiently.
The Challenge
The founders of DutyDingohad identified a persistent pain point in software development agencies. Creating tickets from Slack discussions was a time-consuming and repetitive task, often leading to information loss and context fragmentation. Project managers struggled to capture all relevant information from chat conversations, and teams frequently lost track of previous discussions, resulting in redundant conversations and reduced productivity.
Kickstage's Solution
To address these challenges, we developed a sophisticated AI-powered Slack chatbot. This innovative tool automatically creates tasks from Slack conversations, seamlessly integrates with popular project management platforms like Trello, Jira, Asana, and ClickUp, and maintains context from past conversations. It also efficiently handles multimedia content attachments, ensuring no crucial information is lost in the process. The technology stack behind this solution is as impressive as its functionality. The development team leveraged DSPy for prompt optimization, LLAMA 3 for local model execution to ensure data privacy, and Kubernetes for scalable multi-tenant deployments. The team also experimented with pgVector and Qdrant for efficient vector database solutions. The backend, written in TypeScript and running on Node.js, uses Supabase as a reliable, vendor-agnostic database system, future-proofing the architecture for potential scaling and evolution.
Technical Challenges and Solutions
One of the primary hurdles we faced was optimizing prompts to generate consistent results in the format customersexpected. The initial trial-and-error approach had its limitations, which led to the adoption of DSPy, a framework foralgorithmically optimizing language model prompts and weights. This tool allowed us to break down complex problems, fine-tune prompts for optimal performance, and adapt quickly to changes in the pipeline, language model, or data without extensive manual adjustments.
Data privacy emerged as another critical concern during development. Initially using public models like OpenAI's GPT-4 and Anthropic's Claude, we quickly recognized the need for a more secure solution. In partnership with Hostzero, we implemented a scalable, self-hosted approach using the open-source LLAMA 3 model on Kubernetes. This pivot ensured that customer data never leaves their servers while maintaining the high-quality results expected from public models.
Open AI
DSPy
Llama
Postgres
Qdrant
Supabase
Node.JS
TypeScript
Results and Impact
The impact of our solution for DutyDingo has been substantial. After a year of operation, clients reported significantimprovements in their software development process. Development teams experienced reduced context switching and improved accuracy in ticket creation, leading to up to 5% faster product shipping times. On average, clients saved 8 man-hours per project monthly. Perhaps most importantly, developer satisfaction noticeably increased due to theimproved workflow and reduced administrative overhead.
Working with the DutyDingo team on this cutting-edge technology was an incredible opportunity to deliver real value to clients. We were so impressed with the product that we immediately integrated it into our own internal technology stack.
Conclusion
Through this project, our teams demonstrated their expertise in AI, cloud technologies, and software integration,transforming a common industry pain point into an innovative, time-saving solution. By overcoming significant technical challenges in prompt optimization and data privacy, we delivered a robust, scalable product that not only met but exceeded client expectations. This case study showcases our ability to tackle complex problems and deliver solutions that provide tangible, measurable benefits to clients in the software development industry.