- AI-powered corporate chatbot for internal IT support in insurance
- Helps employees and agents work with core insurance systems (policy admin, claims, billing, CRM, etc.)
- Answers “how-to” questions using existing IT system help, manuals, and FAQs
- Human-in-the-loop: seamless escalation to IT support operators for complex or missing-case queries
- RAG-based knowledge layer built from IT help portal and technical documentation
- Continuous quality evaluation of answers (accuracy, completeness, applicability)
- Tech stack: TypeScript, Node.js, MongoDB, LangChain (RAG)
- Project timeline: 2025, team of 2 developers
The client is a group of Ukrainian insurance companies operating nationwide under a shared IT landscape.
They manage multiple product lines (motor, property, health, travel, corporate risks) and rely on several core systems for policy administration, claims processing, and partner / agent interactions.
The group operates under a non-disclosure agreement (NDA), so specific brand names and figures remain confidential. The organization includes:
- Central offices and regional branches
- Large agent and broker networks
- Call centers and back-office teams relying daily on the same IT platforms
- High volume of repetitive IT questions: “Where do I change the policyholder address?”, “How to register a claim for this product type?”, “Why is this button disabled?”
- Scattered help content across different places: built-in system help, separate help portal, PDF manuals, training slides, emails from IT, etc.
- Long response times and context switching for the IT support team, especially during product launches, campaigns, or regulatory changes.
- Inconsistent guidance: different support agents sometimes gave different instructions for the same scenario or system version.
- Difficult onboarding: new employees and agents needed significant “hand-holding” to learn systems and processes, slowing time-to-productivity.
- AI chatbot in a corporate messenger (e.g., Telegram) as a single entry point for IT system “how-to” questions from employees and agents.
- RAG-based knowledge layer built on existing IT help content (online help, help portal, manuals, FAQs, troubleshooting guides).
- Automatic answers for standard, documented scenarios with step-by-step guidance grounded in retrieved help articles.
- Human-in-the-loop escalation to IT support operators in Telegram for complex, unclear, or undocumented cases, with full conversation context passed through.
- Storage and vectorization of validated operator answers so the chatbot can reuse proven solutions for similar future questions.
- Continuous quality evaluation of AI and operator answers (correctness, applicability, clarity, tone) feeding into prompt and retrieval tuning.
- Scheduled synchronization and re-indexing of IT help content to keep the knowledge base aligned with current system versions and processes.
- Backend implemented in TypeScript / Node.js with MongoDB, LangChain-based RAG pipelines, and integrations with the Telegram Bot API and IT help/documentation systems.
Reduced load on IT support and system trainers
Many standard “how-to” questions are handled automatically, allowing IT staff to focus on incidents, system improvements, and complex cross-system issues.
Faster answers for employees and agents
Front-line staff get immediate, actionable guidance inside the messenger they already use, without searching across multiple help sources or waiting in queues.
More consistent system usage
Answers are grounded in the same, up-to-date IT help content, reducing variability in how processes are executed and improving data quality in core systems.
Shorter onboarding time
New hires and agents can learn the systems by asking natural-language questions and receiving guided, step-by-step instructions, reducing dependence on classroom training.
Insight into training and documentation gaps
Analytics on repeated questions and escalation reasons highlight where help content is missing, unclear, or outdated, helping IT and business teams improve documentation and system UX.
- 2 Developers