π§© 1. Why This Matters: The Problem We Solved
Growing companies with extensive internal documentation often face a common challenge: employees spend too much time searching for information or asking repetitive questions. This reduces productivity and increases the load on support teams.
Typical scenarios:
- A new employee canβt find the contract template;
- A manager forgets how to issue an invoice;
- A developer is unsure who should review their code.
π οΈ 2. Solution Architecture Overview
Before diving into the details, here's an overview of our architecture, which includes:
- RAG (Retrieval-Augmented Generation) β a hybrid AI approach combining retrieval and generation;
- Vector storage β for semantically meaningful document retrieval;
- Telegram bot β as the user interface for employees;
- Redmine API β to automate task management and time tracking;
- GitLab β as the living documentation source.
π§ 3. Preparing the Knowledge Base and Vectorization
The companyβs internal knowledge was stored as .md
files in GitLab. We:
- Automated document extraction;
- Cleaned and preprocessed the content;
- Vectorized the content using
sentence-transformers
(Learn more); - Indexed the content in Elasticsearch.
π Vector Databases Overview
π€ 4. Telegram Bot β Employee Interface
Employees interact with the bot by asking questions directly in chats:
- "How do I log time for Project Alpha?"
- "Which NDA template should I use?"
If the bot is unsure of the answer, it forwards the question to a support channel.
π 5. Self-Learning and Answer Quality Management
Every unanswered question signals a documentation gap.
- We log "unanswered" queries;
- Manually review and tag them;
- Create or update documents;
- Re-index the knowledge base.
π RAG Quality Evaluation β OpenAI Cookbook
π§± 6. Redmine Integration
With Redmine API integration, employees can:
- View their assigned tasks;
- Log time directly via the bot;
- Generate invoices;
- Initiate contract preparation.
π Redmine API Docs
π¬ 7. Example Queries
Examples:
π§: How do I issue an invoice for March?
π€: Here's the instruction: [link]
π§: Who should review my pull request?
π€: Your team lead or a senior developer. More info: [link]
π 8. Outcomes and Scalability
Results:
- Reduced support load;
- Faster onboarding for new employees;
- Routine processes automated.
Scalability options:
- Integration with Slack, MS Teams;
- Multi-language support;
- Integration with other APIs (CRM, ERP, etc.).
π€ Want Similar Results?
If you'd like to automate your internal knowledge access and streamline business processes, get in touch with us. Let's build it together.