Andres Berrios@andresberrios

Objective: Address the needs of marketing professionals for a succinct, efficient, and effective CV Profile - a concise encapsulation of their work activities and workstyle. Utilize AI21's generative AI technology for this task. Develop a unique prompt using a chain-of-thought approach, allowing marketing professionals to quickly create a personalized CV profile. This solution streamlines the updating process of a CV profile. As professionals grow and their skill set diversifies, they can edit their work activities and work style, paste the prompt, and swiftly generate an updated profile. The key advantage of our solution is speed and efficiency. It removes the tedious process of data sourcing, cleaning, and repetitive trial-and-error prompt creation, providing a high-quality output quickly.

Whatever work you do, there is one question that all of us face: How will AI impact my job?! Societally, we are not answering this question very well. The majority of messages are fear-based and vague. Our objective was to figure out how to focus on any job and recognize tasks that can be automated and augmented by generative AI. Within the machine learning workflow, we focused on prompt design, hyperparameter tuning, training and experiment tracking. Though our knowledge harvesting team didn’t include any software engineers, we found that the Generative AI Studio of Google Vertex AI was a powerful way to achieve No-Code AI. So, flip the paradigm! Don’t dwell on data, notebooks, and APIs! Prompt well!

Egregore: An implementation of LangChain and Claude for Accelerated Competitive Analysis Our Knowledge Harvesting team aimed to create an AI solution for a crucial stage of the product development process: Analyzing competition. Understanding the competition is vital to decipher a market and assess market attractiveness. Accurate knowledge of the competition helps product and marketing managers communicate product functions, features, and benefits effectively. We explored techniques to structure information for LangChain and opted to use prompts, testing them in GPT-4 and Claude 2. Additionally, we experimented with semantic triples to drive agent behaviors. Triples and knowledge graphs are powerful and interpretable ways to structure and curate data. In a real-world application, we focused on five ultrasound devices, enabling rich comparisons, including regulatory requirements. We delved into FDA processes for medical devices. Though we didn't fully structure this complex data, we captured the necessary elements to make it AI-digestible. With a 55-page Integrated Product Management Process as reference, we converted its step-by-step guidance into prompts and agents. While we encountered some uncompleted tests to assess AI configurations, we successfully developed workflows, data structures, and prompts to leverage LangChain's potential. Although more work lies ahead, we established a strong foundation for further exploration. Our team's experience in using AI for a real-world scenario was good! In just 3 days, we covered significant ground, demonstrating the tremendous potential of AI to augment human intelligence. Our project's ambition affirmed the possibilities of utilizing AI to enhance product management and innovation.