Our LLM deveelopment team chose Meta Llama-3, along with open-source models for flexible, cost-efficient language understanding and automated decision extraction.
Launch Enterprise-Grade Microservices in Days - Get Your Free Boilerplate
About the Client
Brunswick is a mid-to-large consulting firm with a globally distributed executive team working across North America, Europe, and Asia. They operate with multiple business units overseen by a central executive panel that meets mostly via virtual meetings and sessions.
Country
USA
Industry
Management and Operations
Team Size
5+ Resources
Hours Invested
9000+
Problem Statement
Executive teams spend hours in meetings, yet key decisions often get lost in transcripts, and action items slip through the cracks. As valuable insights stay buried in conversations, teams struggle with follow-ups. Our challenge was to extract data and insights from those panel discussions that company leaders can act on without adding more meeting overhead.
Project Overview
The project began with a straightforward request from the management group - help us extract the most value from every meeting. To address this challenge, we turned to AI. Using a fine-Meta Llama 3 model, the AI meeting assistant listens to meetings, transcribes conversations in real time, and pulls out key decisions, action items, and priorities.
Radixweb’s AI development team plugged the system directly into the tools they use. For example, we set up the AI meeting intelligence platform with Asana to feed task trackers and project boards, with Notion to feed documentation, and into Microsoft Teams to feed conversation threads. As the solution eliminates the need to summarize and track follow-ups manually, Brunswick executives can see key decisions, assigned owners, and next steps immediately.
Client Remarks
It works like a personal assistant but way more efficient and accurate. Of course it’s because of AI, yet the difference is made by the team building it. We’d explored similar solutions for some time, but the proposals we got just didn’t click until Radix came in. Now our meetings actually feel productive.
Jayne Thomas-Daoulas
VP Strategy, Brunswick, Inc.
There was no margin for error. Every piece of data had to be correct, so we made sure only the most experienced team handled it, through multiple validation rounds and keeping a close eye on every detail. That attention to detail made all the difference.
Rahul Shrimali
Project Lead at Radixweb
Risks and Challenges
- Output Accuracy - Capturing detailed conversations without misinterpreting decisions, priorities, speaker, or context. It should avoid false positives or incorrect action items.
- Real-Time Processing - Transcribing, diarizing, parsing, and analyzing audio quickly enough to provide immediate insights during or right after meetings.
- Speaker Attribution - Correctly identifying who said what, especially in multi-speaker discussions with overlapping speech.
- Contextual Understanding - Linking discussions to ongoing projects, previous decisions, and overall meeting goals. The AI meeting summary tool must understand the difference between productive talk and casual chat.
- Data Security & Privacy - Handling private executive-level conversations and complying with internal security policies and enterprise protocols.
Solutions Delivered
Accurate Meeting Notes, Summaries, and Alerts
The executive meeting analytics platform creates organized notes with timestamps, key outcomes, and alerts for missed items in multiple meetings or overlapping discussions.
Real-Time Transcription and Follow-Up Metrics
It uses live transcription with speaker tagging and sentiment tracking to automatically quantify meeting efficiency through metrics like action completion rate and decision turnaround time.
Speaker Identification & Context Tracking
Voice differentiation and NLP-based entity recognition are the two main techniques our AI developers implemented to tag speakers, track topic continuity, and maintain context across recurring executive sessions.
Automated Action Item Extraction
The LLM-powered meeting summary platform identifies follow-ups, owners, assignees, and due dates using intent and dependency detection models. It syncs structured actions directly with task systems for accountability.
Context-Aware Meeting Memory
To allow the language model to refer to prior meetings when generating summaries or recommendations, the platform retains historical discussion data, notes, decisions, and dependencies.
Decision Intelligence Layer
Highlighting critical decisions, their reasoning, and projected impact is another standout feature of the enterprise meeting AI solution. With trackable records, it integrates into executive dashboards.
Build and Deploy AI Models
With guaranteed delivery timelines, success KPIs, and post-deployment maintenance.
Technology Stack
LLM Integration
Cloud Infrastructure
AWS S3 was our first choice for a scalable, secure storage of transcripts, meeting data, and analytics. It also supports real-time AI processing and long-term historical data retention.
Workflow Automation
We used AWS SES to automate email notifications, follow-ups, reminders, and task assignments. Action items from meetings are sent directly into Teams channels.
Client Benefits
For over a year, our AI/ML specializing experts were designing the Llama-powered AI meeting platform on AWS. The end solution had noticeable improvements in follow-up time, task completion rate, and enterprise decision making.
Significant Reduction in Meeting Follow-Up Time
Follow-ups now happen roughly 3X faster than before. Executives can review key decisions and next steps immediately. The LLM meeting insights platform has cut hours of post-meeting admin each month.
Faster Action Item Completion Rates
Teams now save 10+ hours every week in task completion when action items are automatically captured, assigned, and synced to Asana. Follow-throughs are tracked in real time.
Enhanced Institutional Memory of Key Decisions
Decisions and discussion context of meetings are preserved in a timely manner. With the AI platform for executive meeting insights, the management team gets instant access to years of historical insights and avoids redundant topics.
Complete Privacy of Sensitive Conversations
97%+ users say the system helps them stay fully present during meetings, confident that all conversations remain encrypted and under strict access controls. AI processing never exposes information externally.
Not Ready for Full-Scale Adoption?
Start with a time-boxed POC. We deliver a working prototype and TAT guarantee before you commit.