Charting AI Frontiers: Microsoft + Madrona Ventures GenAI Hackathon

Microsoft Personal Capital and International Unbiased Software program Vendor Groups, in collaboration with Madrona Ventures, just lately organized a GenAI hackathon for portfolio firms, hosted on the Redmond campus. 33 members from Amperity, Clari, Highspot, and SeekOut, fashioned seven groups, to discover generative AI options. Previous to the Hackathon, members crafted distinctive drawback statements and strategized with Microsoft coaches on methods to leverage Generative AI’s potential to deal with enterprise or finish buyer challenges, improve product options, or propel product/characteristic growth. The objective was to finish the two-day hackathon with proof of ideas, AI-enabled, and a path ahead to take these options to manufacturing.

The groups have been judged throughout three key themes: 1) the use case and innovation, 2) putting people on the core with a concentrate on the answer’s worth to people, and three) the incorporation of Accountable AI practices.

Let’s see what they got here up with:

Amperity

Amperity, a Buyer Information Platform, empowers manufacturers to develop by gaining a deep understanding of their clients. It gives an information basis for buyer acquisition, retention, personalised experiences, and privateness administration.

The Amperity group determined to leverage AI to make it simpler for advertising management and no-code enterprise resolution makers to work together with knowledge from disparate, unstructured knowledge unfold throughout emails, social media, and transactions, to raised plan and serve their clients.

The undertaking centered on utilizing GPT-4 via Azure Open AI Service to develop a conversational assistant to investigate buyer knowledge and allow all group members to make data-driven choices. They reworked conversations on to SQL queries executed in opposition to buyer profiles databases and displayed in tables. This course of simplifies knowledge interpretation, making it accessible to all staff.

Amperity then used OpenAI’s code interpreter to Generate graphical representations.

The result’s a unified, 360-degree view of a buyer that was beforehand scattered throughout a number of touchpoints. This holistic perspective permits entrepreneurs, buyer expertise groups, and decision-makers to know their clients on a deeper degree.

One other means Amperity creatively used AI is showcased via icons from question outcomes:

The crown icon is chosen by the AI mannequin based mostly on the generated output “PLATINUM_CUSTOMERS”.

The format, icons, and general presentation can now be dynamically generated. This multifaceted method, the place the aspect is chosen by AI, represents a brand new period in UI design, opening a world of potentialities for startups to create extra adaptive and user-friendly interfaces.

A significant spotlight of the undertaking addressed potential points with SQL by passing any SQL errors again to GPT for auto-correction. This saved time and improved the system’s effectivity and reliability. Amplified demonstrated a degree of automated problem-solving, which is a game-changer for startups seeking to develop shortly.

Highspot

Highspot is a gross sales enablement platform that equips gross sales groups with the instruments, content material, and insights wanted to enhance efficiency and drive income.

They offered three approaches to growing gross sales productiveness via AI.

Highspot Crew 1

Showcased their revolutionary use of AI to enhance consumer expertise and enhance productiveness.

Highspot Crew 1 centered on three main themes: automation recommendation, co-creation, and a characteristic they’re tentatively calling, “Ask Something”. The corporate has applied AI in its product to generate descriptions for content material, create assembly summaries, and compose messages, amongst different options.

‘Ask Something’ surfaces paperwork and content material based mostly on questions posed by the consumer. This characteristic may very well be particularly helpful for customers in enablement and advertising to generate descriptions for content material.

They demonstrated that whereas trying to find a doc they’ll present solutions to a consumer’s particular questions in regards to the doc itself. They even automated follow-ups, the place the system can ship follow-up emails that embrace particular particulars mentioned within the rep’s buyer assembly.

Highspot is engaged on integrating these functionalities into their extensions for Slack and Microsoft Groups. This may allow customers to ask questions inside these shoppers and obtain direct solutions in regards to the product or some other associated question.

Their implementation works significantly effectively for giant paperwork the place the search can recall particulars deep throughout the textual content.

For recordings in Microsoft Groups Highspot can ask, “What did they speak about once they talked about matter X?” or “What did particular person XYZ say a few sure matter?”

In essence, making it simpler for gross sales groups to entry necessary info shortly.

The Highspot group utilized gpt-35-turbo-16k from OpenAI, mixed with Microsoft applied sciences, to reinforce their pure language understanding and technology capabilities. Utilizing Azure OpenAI’s sturdy infrastructure and complete toolset enabled Highspot to develop, check, and deploy their AI options swiftly and effectively.

The important thing constructing blocks of Highspot’s structure embrace OpenAI Perform routing for question classification and knowledge binding, and dynamic UI rendering based mostly on question intent. These parts work in tandem to create a seamless consumer expertise, the place the system can perceive consumer queries, classify them, and render applicable responses dynamically.

Highspot Crew 2

Highspot Crew 2 centered their efforts on integrating gross sales enablement capabilities extending Copilot for Gross sales. The undertaking goals to reinforce the productiveness of gross sales groups by streamlining entry to related knowledge, decreasing the necessity for context switching between completely different instruments and by offering related insights and automating duties all inside Outlook.

The context {that a} vendor must reply an electronic mail is pulled from the Highspot’s platform into Copilot. This consists of views, downloads, recipients from Highspot in addition to CRM knowledge from Salesforce.

Leveraging Salesforce knowledge permits the Copilot to supply predictive content material suggestions. By analyzing the CRM knowledge, Highspot can establish developments and patterns in buyer habits and supply gross sales reps with probably the most related content material and alternatives for every distinctive buyer interplay.

This permits the sellers to make choices, streamline their workflows, and finally shut offers quicker.

The group utilized Microsoft Copilot for Gross sales with Highspot Connector to create a dynamic Copilot UI that adjusts based mostly on the Highspot context. This integration allowed for a seamless transition between Highspot and Copilot, enhancing the consumer expertise and boosting productiveness.

Highspot Crew 3

Highspot Crew 3 centered on teaching gross sales representatives and showcased why it’s so important for sellers. Drawing from educational analysis, they highlighted the numerous enchancment in efficiency when college students obtain one-on-one mentorship or teaching. In a company context, the identical precept applies. The truth is, gross sales reps who obtain two hours or extra of teaching per week have 56% or larger win charges.

Teaching typically falls by the wayside in lots of organizations resulting from time constraints and an absence of formalized teaching strategies. Crew 2 aimed to provide gross sales reps suggestions on their abilities demonstrated in real-world conferences and help managers in offering personalised teaching.

To attain this, they utilized Microsoft Groups’ transcription and recording API, together with rubrics, query context, reply guides from Highspot, and Azure’s OpenAI GPT-3.5 Turbo mannequin. They staged a gross sales discovery name between Contoso (the vendor) and Fabrikam (the prospect) and used GPT to create a transcript. This supplied the idea for his or her suggestions system.

Gross sales reps submit a recorded assembly for suggestions. The suggestions is predicated on predefined rubrics and gives detailed insights into their efficiency. Gross sales reps can then submit their conferences, together with the AI-Generated suggestions, to their managers for additional overview. Managers even have the choice to Generate AI-assisted suggestions, which they’ll then personalize earlier than submitting it to the rep.

The principle constructing blocks of their structure included a React entrance finish, Highspot’s assembly ingestion leveraging MS Groups APIs, and Azure OpenAI APIs. Nonetheless, the implementation was not with out its challenges. Figuring out the assessed topic within the immediate proved tough, necessitating varied methods to investigate the transcription to make it useful.

SeekOut

SeekOut is a people-first, AI-assisted expertise acquisition and administration platform that gives complete knowledge, uncovers unnoticed candidates, and helps worker progress.

SeekOutTeam 1

SeekOut Crew 1 showcased a prevalent problem within the job market – private profession teaching. The group highlighted the inequitable distribution of profession teaching providers and the prohibitive prices related to them.

Their resolution? “Yoda,” an AI-driven profession coach, set to democratize profession teaching by making it accessible to everybody.

Profession teaching is commonly out of attain for many staff resulting from its excessive price. Human coaches require an intensive understanding of their consumer or mentee’s background, pursuits, values, and profession aspirations, which regularly requires a major quantity of effort and time. Yoda, makes use of generative AI to know a person’s background, abilities, pursuits, and targets, offering personalised suggestions.

The AI-driven profession coach, Yoda, was designed to know the nuances of a number of industries, a activity that will be difficult for a human. It additionally scales throughout all staff inside a company, making it an economical resolution for HR groups.

Yoda makes use of Microsoft Groups app and integrates with Profession Compass, which pulls in knowledge from varied inner and exterior sources to create a complete profile for every consumer. They used GPT 3.5 turbo to generate and parse the chats, analyze jobs, and profile info to supply customers with personalized suggestions.

It solves two particular use circumstances: 1) personalised profession pathing and scanning the whole group for brand spanking new alternatives, and a pair of) ideas of inner networking contacts and potential mentors. Yoda considers the consumer’s profile, transition knowledge, and job positions to recommend potential roles that the consumer would possibly excel in.

Contemplate the consumer persona, Jeanne, a techniques engineer seeking to transition into software program engineering.

The chatbot integrates with studying techniques and inner job boards, offering customers with a wide range of sources to help their profession development. As well as, it suggests connections throughout the group that customers can attain out to for insights and recommendation.

AI drafts introductory messages for customers to ship to potential connections. This characteristic helps customers break the ice and provoke conversations, which might typically be a frightening activity.

The group didn’t cease at offering suggestions. Yoda constantly scans the group for brand spanking new alternatives that match the consumer’s profession targets. When an acceptable alternative arises, the chatbot sends a customized message to the consumer, summarizing the place and explaining why it is likely to be a superb match.

A consumer can save their profession selection via the chat utilizing pure language, integrating actions along with pure content material technology and suggestion.

The SeekOut Crew 1 believes Yoda affords a singular resolution for hyper-personalized profession teaching and scalability. Their presentation included the potential for this software to supply high-level insights to HR and organizational leaders, similar to widespread profession progressions, desired abilities, and developmental wants.

They plan to additional combine the Profession Compass app inside groups, gamify abilities growth, and add Yoda to the supervisor’s expertise to assist managers in serving to staff develop and develop.

SeekOutTeam 2

SeekOut Crew 2 showcased a game-changing resolution to allow expertise leaders to reply essential questions like, “What abilities does my group have and the way can I successfully discover folks with these abilities and manage them higher?” The group launched a dynamic abilities platform that leverages Microsoft 365 and OpenAI to supply insights into the skillsets of staff, providing a brand new degree of understanding for expertise leaders.

Understanding Worker Expertise

Crew 2’s resolution centered on serving to organizations perceive the expertise they’ve and methods to develop it. They developed a abilities intelligence platform that ingests knowledge from varied sources, together with resumes, GitHub, and different skilled paperwork.

The platform makes use of this knowledge to construct relationships between abilities and the way they relate to the bigger organizational image. It additionally gives proof of how completely different staff have demonstrated their abilities.

The dynamic abilities platform seeks to know what the worker has accomplished and how much abilities they’ve demonstrated of their productiveness instruments and resumes. The platform is designed to adapt to real-time context when somebody is utilizing it, permitting for extra particular queries. For instance, as a substitute of asking which individuals have machine studying as a talent, a expertise chief can ask, “How is machine studying utilized by knowledge scientists at my firm?”

The platform’s integration with Microsoft 365 utilizing Microsoft Graph API provides extra depth to the abilities platform. It takes paperwork from Microsoft 365, feeds them into Azure OpenAI to generate info on the abilities, after which provides this to the abilities database. This info is then accessible via the Expertise Explorer software.

The group demonstrated how a hiring supervisor at Firm X might use the platform to know the abilities of a program supervisor. The platform breaks down the abilities profile for the function, exhibiting useful, technical, and management abilities. The supervisor can then delve deeper into every talent, seeing proof from resumes, job postings, and Microsoft 365 recordsdata.

The platform additionally gives a means for the supervisor to see how particular abilities, like undertaking administration, are utilized by program managers at Firm X. The supervisor can see paperwork linked to the talent and the small print of those paperwork. This info might help the supervisor perceive what abilities they want for a brand new function or craft a job description for an open place.

Privateness and Subsequent Steps

The SeekOut group took privateness into consideration, making certain all knowledge ingestion runs within the buyer’s atmosphere. There’s a layer of information separation to guard consumer knowledge, and buyer knowledge by no means involves the SeekOut group.

Tech stack

The group utilized the gpt-3.5-turbo mannequin with Microsoft Graph API, Azure OpenAI, and Azure SQL for his or her dynamic abilities platform. The structure of their resolution comprised a number of key constructing blocks. These included the Graph API to obtain group content material from Microsoft 365, an offline OpenAI knowledge pipeline to find out the abilities demonstrated by the authors of the One Drive paperwork, and a SQL database backend to retailer the connection between abilities and staff.

The group’s resolution concerned pulling knowledge from Google Drive and embedding it into Azure’s Cognitive Search. This allowed them to course of the info and make it searchable for the engineering group.

The group showcased their resolution by asking it questions on particular incidents or points. The system recognized related paperwork and supplied hyperlinks to the unique runbooks or incident experiences and any technical documentation in regards to the instruments they’re utilizing.

The group’s undertaking was a captivating demonstration of how AI and cloud applied sciences could be leveraged to streamline inner processes and enhance effectivity. They used GPT 3.5 Turbo, obtainable via Azure OpenAI Service. The structure was based mostly on an ordinary RAG sample applied utilizing Azure Cognitive Seek for the Vector Database.

The answer concerned a number of parts, similar to Google Drive integration for doc ingestion, LlamaIndex for doc processing and orchestration, Azure Cognitive Seek for storing embedded chunks, and Streamlit for a easy chat interface.

Regardless of some challenges, they demonstrated how startups can leverage these to reinforce their inner processes. The group’s dedication to enhancing their system, similar to their plans to deal with the problem with acronyms and codes in semantic search and to collect consumer suggestions, underscores the continual evolution and potential of AI options within the enterprise world.

And the winner was….

Highspot Crew 2! Highspot Crew 2’s resolution not solely aligned with the hackathon’s targets but in addition demonstrated a sensible and impactful method. Because of this, the work of their larger group was included in Satya Nadella’s keynote presentation at Microsoft Ignite.

To our winners, you’ve rocked this along with your unimaginable abilities and innovation. And to all our members, your power and creativity actually lit up this occasion.

Each one in every of you is a star in our tech galaxy. Hold shining and see you on the subsequent hackathon! 🚀

Tags: AI, Hackathon, Madrona