Microsoft Private Capital and Global Independent Software Vendor Teams, in collaboration with Madrona Ventures, recently organized a GenAI hackathon for portfolio companies, hosted at the Redmond campus. 33 participants from Amperity, Clari, Highspot, and SeekOut, formed seven teams, to explore generative AI solutions. Prior to the Hackathon, participants crafted distinctive problem statements and strategized with Microsoft coaches on how to leverage Generative AI’s potential to address business or end customer challenges, enhance product features, or propel product/feature development. The goal was to end the two-day hackathon with proof of concepts, AI-enabled, and a path forward to take these solutions to production.
The teams were judged across three key themes: 1) the use case and innovation, 2) placing humans at the core with a focus on the solution’s value to individuals, and 3) the incorporation of Responsible AI practices.
Let’s see what they came up with:
Amperity
Amperity, a Customer Data Platform, empowers brands to grow by gaining a deep understanding of their customers. It provides a data foundation for customer acquisition, retention, personalized experiences, and privacy management.
The Amperity team decided to leverage AI to make it easier for marketing leadership and no-code business decision makers to interact with data from disparate, unstructured data spread across emails, social media, and transactions, to better plan and serve their customers.
The project centered on using GPT-4 through Azure Open AI Service to develop a conversational assistant to analyze customer data and enable all team members to make data-driven decisions. They transformed conversations directly to SQL queries executed against customer profiles databases and displayed in tables. This process simplifies data interpretation, making it accessible to all employees.
Amperity then used OpenAI’s code interpreter to Generate graphical representations.
The result is a unified, 360-degree view of a customer that was previously scattered across multiple touchpoints. This holistic perspective allows marketers, customer experience teams, and decision-makers to understand their customers on a deeper level.
Another way Amperity creatively used AI is showcased through icons from query results:
The crown icon is selected by the AI model based on the generated output “PLATINUM_CUSTOMERS”.
The layout, icons, and overall presentation can now be dynamically generated. This multifaceted approach, where the facet is selected by AI, represents a new era in UI design, opening a world of possibilities for startups to create more adaptive and user-friendly interfaces.
A major highlight of the project addressed potential issues with SQL by passing any SQL errors back to GPT for auto-correction. This saved time and improved the system’s efficiency and reliability. Amplified demonstrated a level of automated problem-solving, which is a game-changer for startups looking to grow quickly.
Highspot
Highspot is a sales enablement platform that equips sales teams with the tools, content, and insights needed to improve performance and drive revenue.
They presented three approaches to increasing sales productivity through AI.
Highspot Team 1
Showcased their innovative use of AI to improve user experience and boost productivity.
Highspot Team 1 focused on three primary themes: automation advice, co-creation, and a feature they are tentatively calling, “Ask Anything”. The company has implemented AI in its product to generate descriptions for content, create meeting summaries, and compose messages, among other features.
‘Ask Anything’ surfaces documents and content based on questions posed by the user. This feature could be especially useful for users in enablement and marketing to generate descriptions for content.
They demonstrated that while searching for a document they can provide answers to a user’s specific questions about the document itself. They even automated follow-ups, where the system can send follow-up emails that include specific details discussed in the rep’s customer meeting.
Highspot is working on integrating these functionalities into their extensions for Slack and Microsoft Teams. This would enable users to ask questions within these clients and receive direct answers about the product or any other related query.
Their implementation works particularly well for large documents where the search can recall details deep within the text.
For recordings in Microsoft Teams Highspot can ask, “What did they talk about when they mentioned topic X?” or “What did person XYZ say about a certain topic?”
In essence, making it easier for sales teams to access important information quickly.
The Highspot team utilized gpt-35-turbo-16k from OpenAI, combined with Microsoft technologies, to enhance their natural language understanding and generation capabilities. Using Azure OpenAI’s robust infrastructure and comprehensive toolset enabled Highspot to develop, test, and deploy their AI features swiftly and efficiently.
The key building blocks of Highspot’s architecture include OpenAI Function routing for query classification and data binding, and dynamic UI rendering based on query intent. These elements work in tandem to create a seamless user experience, where the system can understand user queries, classify them, and render appropriate responses dynamically.
Highspot Team 2
Highspot Team 2 focused their efforts on integrating sales enablement capabilities extending Copilot for Sales. The project aims to enhance the productivity of sales teams by streamlining access to relevant data, reducing the need for context switching between different tools and by providing relevant insights and automating tasks all within Outlook.
The context that a seller needs to answer an email is pulled from the Highspot’s platform into Copilot. This includes views, downloads, recipients from Highspot as well as CRM data from Salesforce.
Leveraging Salesforce data allows the Copilot to provide predictive content recommendations. By analyzing the CRM data, Highspot can identify trends and patterns in customer behavior and provide sales reps with the most relevant content and opportunities for each unique customer interaction.
This enables the sellers to make decisions, streamline their workflows, and ultimately close deals faster.
The team utilized Microsoft Copilot for Sales with Highspot Connector to create a dynamic Copilot UI that adjusts based on the Highspot context. This integration allowed for a seamless transition between Highspot and Copilot, enhancing the user experience and boosting productivity.
Highspot Team 3
Highspot Team 3 focused on coaching sales representatives and showcased why it is so essential for sellers. Drawing from academic research, they highlighted the significant improvement in performance when students receive one-on-one mentorship or coaching. In a corporate context, the same principle applies. In fact, sales reps who receive two hours or more of coaching per week have 56% or higher win rates.
Coaching often falls by the wayside in many organizations due to time constraints and a lack of formalized coaching methods. Team 2 aimed to give sales reps feedback on their skills demonstrated in real-world meetings and assist managers in providing personalized coaching.
To achieve this, they utilized Microsoft Teams’ transcription and recording API, along with rubrics, question context, answer guides from Highspot, and Azure’s OpenAI GPT-3.5 Turbo model. They staged a sales discovery call between Contoso (the seller) and Fabrikam (the prospect) and used GPT to create a transcript. This provided the basis for their feedback system.
Sales reps submit a recorded meeting for feedback. The feedback is based on predefined rubrics and provides detailed insights into their performance. Sales reps can then submit their meetings, along with the AI-Generated feedback, to their managers for further review. Managers also have the option to Generate AI-assisted feedback, which they can then personalize before submitting it to the rep.
The main building blocks of their architecture included a React front end, Highspot’s meeting ingestion leveraging MS Teams APIs, and Azure OpenAI APIs. However, the implementation was not without its challenges. Identifying the assessed subject in the prompt proved difficult, necessitating various ways to analyze the transcription to make it functional.
SeekOut
SeekOut is a people-first, AI-assisted talent acquisition and management platform that provides comprehensive data, uncovers unnoticed candidates, and supports employee growth.
SeekOutTeam 1
SeekOut Team 1 showcased a prevalent challenge in the job market – personal career coaching. The team highlighted the inequitable distribution of career coaching services and the prohibitive costs associated with them.
Their solution? “Yoda,” an AI-driven career coach, set to democratize career coaching by making it accessible to everyone.
Career coaching is often out of reach for most employees due to its high cost. Human coaches require an extensive understanding of their client or mentee’s background, interests, values, and career aspirations, which often requires a significant amount of time and effort. Yoda, uses generative AI to understand an individual’s background, skills, interests, and goals, providing personalized recommendations.
The AI-driven career coach, Yoda, was designed to understand the nuances of multiple industries, a task that would be challenging for a human. It also scales across all employees within an organization, making it a cost-effective solution for HR teams.
Yoda uses Microsoft Teams app and integrates with Career Compass, which pulls in data from various internal and external sources to create a comprehensive profile for each user. They used GPT 3.5 turbo to generate and parse the chats, analyze jobs, and profile information to provide users with customized recommendations.
It solves two specific use cases: 1) personalized career pathing and scanning the entire organization for new opportunities, and 2) suggestions of internal networking contacts and potential mentors. Yoda considers the user’s profile, transition data, and job positions to suggest potential roles that the user might excel in.
Consider the user persona, Jeanne, a systems engineer looking to transition into software engineering.
The chatbot integrates with learning systems and internal job boards, providing users with a variety of resources to aid their career progression. In addition, it suggests connections within the organization that users can reach out to for insights and advice.
AI drafts introductory messages for users to send to potential connections. This feature helps users break the ice and initiate conversations, which can often be a daunting task.
The team didn’t stop at providing recommendations. Yoda continuously scans the organization for new opportunities that match the user’s career goals. When a suitable opportunity arises, the chatbot sends a personalized message to the user, summarizing the position and explaining why it might be a good fit.
A user can save their career choice through the chat using natural language, integrating actions in addition to pure content generation and suggestion.
The SeekOut Team 1 believes Yoda offers a unique solution for hyper-personalized career coaching and scalability. Their presentation included the potential for this tool to provide high-level insights to HR and organizational leaders, such as common career progressions, desired skills, and developmental needs.
They plan to further integrate the Career Compass app within teams, gamify skills development, and add Yoda to the manager’s experience to support managers in helping employees grow and develop.
SeekOutTeam 2
SeekOut Team 2 showcased a game-changing solution to enable talent leaders to answer critical questions like, “What skills does my organization have and how can I effectively find people with those skills and organize them better?” The team introduced a dynamic skills platform that leverages Microsoft 365 and OpenAI to provide insights into the skillsets of employees, offering a new level of understanding for talent leaders.
Understanding Employee Skills
Team 2’s solution focused on helping organizations understand the talent they have and how to grow it. They developed a skills intelligence platform that ingests data from various sources, including resumes, GitHub, and other professional documents.
The platform uses this data to build relationships between skills and how they relate to the larger organizational picture. It also provides evidence of how different employees have demonstrated their skills.
The dynamic skills platform seeks to understand what the employee has done and what kind of skills they have demonstrated in their productivity tools and resumes. The platform is designed to adapt to real-time context when someone is using it, allowing for more specific queries. For example, instead of asking which people have machine learning as a skill, a talent leader can ask, “How is machine learning used by data scientists at my company?”
The platform’s integration with Microsoft 365 using Microsoft Graph API adds more depth to the skills platform. It takes documents from Microsoft 365, feeds them into Azure OpenAI to generate information on the skills, and then adds this to the skills database. This information is then accessible through the Skills Explorer application.
The team demonstrated how a hiring manager at Company X could use the platform to understand the skills of a program manager. The platform breaks down the skills profile for the role, showing functional, technical, and leadership skills. The manager can then delve deeper into each skill, seeing evidence from resumes, job postings, and Microsoft 365 files.
The platform also provides a way for the manager to see how specific skills, like project management, are used by program managers at Company X. The manager can see documents linked to the skill and the details of these documents. This information can help the manager understand what skills they need for a new role or craft a job description for an open position.
Privacy and Next Steps
The SeekOut team took privacy into account, ensuring all data ingestion runs in the customer’s environment. There is a layer of data separation to protect user data, and customer data never comes to the SeekOut team.
Tech stack
The team utilized the gpt-3.5-turbo model with Microsoft Graph API, Azure OpenAI, and Azure SQL for their dynamic skills platform. The architecture of their solution comprised several key building blocks. These included the Graph API to download team content from Microsoft 365, an offline OpenAI data pipeline to determine the skills demonstrated by the authors of the One Drive documents, and a SQL database backend to store the relationship between skills and employees.
The team’s solution involved pulling data from Google Drive and embedding it into Azure’s Cognitive Search. This allowed them to process the data and make it searchable for the engineering team.
The team showcased their solution by asking it questions about specific incidents or issues. The system identified relevant documents and provided hyperlinks to the original runbooks or incident reports and any technical documentation about the tools they are using.
The team’s project was a fascinating demonstration of how AI and cloud technologies can be leveraged to streamline internal processes and improve efficiency. They used GPT 3.5 Turbo, available through Azure OpenAI Service. The architecture was based on a standard RAG pattern implemented using Azure Cognitive Search for the Vector Database.
The solution involved several components, such as Google Drive integration for document ingestion, LlamaIndex for document processing and orchestration, Azure Cognitive Search for storing embedded chunks, and Streamlit for a simple chat interface.
Despite some challenges, they demonstrated how startups can leverage these to enhance their internal processes. The team’s commitment to improving their system, such as their plans to address the issue with acronyms and codes in semantic search and to gather user feedback, underscores the continuous evolution and potential of AI solutions in the business world.
And the winner was….
Highspot Team 2! Highspot Team 2’s solution not only aligned with the hackathon’s goals but also demonstrated a practical and impactful approach. As a result, the work of their greater team was included in Satya Nadella’s keynote presentation at Microsoft Ignite.
To our winners, you’ve rocked this with your incredible skills and innovation. And to all our participants, your energy and creativity truly lit up this event.
Every one of you is a star in our tech galaxy. Keep shining and see you at the next hackathon! 🚀
