The project was implemented for VisionSystems, a company that develops solutions in the field of business automation and knowledge management.
Secretopus is a corporate communications platform that helps companies collect, structure, and use internal knowledge. The solution is presented as a universal tool for building a knowledge base and automating team communications.
The LighTech team participated in the creation of a system that transforms disparate communications within the company (meetings, correspondence, discussions) into a single, secure and accessible knowledge base.
Basic system requirements:
Reliable storage and processing of confidential data, taking into account corporate security standards.
Support for automatic recording and sammarification of video meetings without user involvement with integration into popular video conferencing systems (Zoom, MS Teams, Google Meet).
Extract key insights from communications and automatically generate a knowledge base based on meetings and discussions.
The ability to customize and connect your own AI agents to the company's tasks.
The possibility of both SaaS deployment and installation inside the client's infrastructure.
As part of the project, our team worked on the implementation of a number of technological components of the platform.:
Integration with the corporate IT infrastructure. Collecting information from messengers, video calls, and mail. Support for MS Teams, Zoom, Google Meet. At the same time, security standards are strictly observed: data is encrypted, transmitted through secure channels, and processed in isolated environments.
Building a knowledge base through the RAG model. The implementation of an intelligent mechanism that transforms texts from meetings and correspondence into interconnected blocks of knowledge. This is the so-called knowledge graph, which makes it easy to find the information you need.
Architecture of AI agents. Creating a core running on intelligent assistants (agents). They can analyze meetings, prompt in real time, generate summaries and reminders. We have provided an opportunity to connect our own company agents and customize them for tasks.
Flexible access control. An interface has been developed that allows administrators to manage access rights, configure channels for commands, and synchronize data with calendars and other services.
Automatic documentation and offline access. Each meeting is recorded, sammarialized, and stored in a centralized repository. Access is possible even in offline mode.
The solution is implemented in such a way as to eliminate manual labor and fragmentation of information. All materials are stored in a centralized repository, which is accessed via a user-friendly web interface. After the meeting ends, the system automatically generates a Sammari with the help of an AI assistant, including basic solutions, announced tasks and agreements between the participants.
Research and analytics
We started by exploring how communications actually work in distributed IT teams. We immersed ourselves in daily scenarios — who interacts with whom, how agreements are fixed, where and why important information is lost. As a result, the main problems have been revealed: everything is fragmented, tasks are "lost along the way," and after rallies there is often no common understanding of what to do next.
UX prototyping
Based on these observations, we have designed interfaces that help you quickly find what you need and not lose the context. Focus on the moments when you need to act here and now. Minimum clicks, maximum benefits: the necessary information is at hand, solutions are on the surface.
UI design
A concise UI kit has been developed, adapted for professional use in the IT environment. Light and dark themes are supported, and the color palette focuses on calm shades with accents for visual labels and insights.
Integration with AI
Automatic conference recording
The key technical component is the development of a script running in a Docker environment, with the ability to connect to Google Meet, Zoom and Microsoft Teams. It provides:
automatically connect and record meetings with video and audio (via FFMPEG);
accepting links via HTTP endpoint;
flexible authorization and recording permission settings;
easy access to files via the built-in HTTP server.
Security and deployment
The entire system is deployed in the customer's private infrastructure (AWS or on-premise). We have ensured the complete autonomy of the solution — without accessing third-party APIs and storing all data inside the secure client contour.
Testing and piloting
Secretopus is a live tool that helps teams work with information before, during, and after communication. Below are the main features of the system.:
Automatic summary of meetings and assignment of tasks. The AI agent captures the key points of the meeting, forms a brief sammari, insights on topics, extended notes and a complete decoding without distortion. Based on the notes, it automatically creates tasks with responsible people and deadlines.
Navigating the knowledge graph. Building a knowledge map where all information is connected logically. The AI assistant searches not by keywords, but by meaning - it finds what is needed in the context of previous discussions and the knowledge base.
Structured and secure storage. All discussions are archived and accessible via search, chat, or API. If necessary, storage in a closed corporate circuit.
Flexible access and workspace system. The spaces are divided into teams and projects. Access rights are configured individually (who sees, who edits, who just receives notifications).
Real-time help. The AI agent is present at meetings, records agreements, suggests actions and reminders.
Data-based analytics and recommendations. AI analyzes the content of meetings, forms recommendations, collects information on the topic for the period and initiates actions — without user participation.
Integration with corporate tools. The service connects to Jira, Confluence, Google Calendar and other tools. Tasks and documents are created automatically, without manual input.
The solution is based on a developed script that functions inside a Docker container and provides recording of video and audio streams using ffmpeg. He receives links to meetings via HTTP endpoint, connects to them automatically and keeps a record until completion.
Authorization on the platforms is implemented through configuration files mounted during container creation, which is especially important for platforms with additional authentication. The entire configuration is set in a yaml file, which simplifies the deployment of the solution in any infrastructure.
During the development of Secretopus, special attention was paid to the technological stability and flexibility of the solution. The system architecture is based on the following approaches:
Local processing and autonomy. All data is processed inside the customer's IT infrastructure, which eliminates external dependencies and meets high security requirements.
Modularity and extensibility. The system was originally designed as a set of interchangeable modules. This makes it easy to integrate your own AI agents and specialized handlers.
Horizontal scaling. The platform supports working with thousands of users simultaneously, without loss of response speed or processing quality.
The Secretopus platform has entered a production-ready state. Its implementation takes up to 2 weeks if the solution is installed in the company's contour and in a matter of minutes if the tool is used as a SaaS.
Already at the pilot stage, employees note the simplification of work meetings, more accurate fixation of agreements and preservation of context. An AI agent takes over the routine work of analyzing meetings and helps reduce information loss.