AI Consultant Widget for a Training & Consulting Platform
Challenge
A training and consulting company operating two platforms — Kadroland (HR) and 7eminar (accounting) — built their reputation on expert consultations, but human consultants had a hard cap on capacity. Availability was limited to business hours, peak periods caused delays, and scaling meant hiring more specialists.
The client needed a way to serve a growing user base without proportionally growing the team or compromising the quality their audience expected.
A training and consulting company operating two platforms — Kadroland (HR) and 7eminar (accounting) — built their reputation on expert consultations, but human consultants had a hard cap on capacity. Availability was limited to business hours, peak periods caused delays, and scaling meant hiring more specialists.
The client needed a way to serve a growing user base without proportionally growing the team or compromising the quality their audience expected.
Solution
Brights built an autonomous AI consultant widget from scratch — no existing infrastructure, no off-the-shelf chatbot.
The process started with a PoC: a Telegram bot tested against 250 real expert questions in HR and tax law, scored by internal specialists. With 80% of answers rated excellent, the concept was validated and MVP development began.
The final solution features a multi-agent architecture for specialized query handling, RAG (Retrieval-Augmented Generation) for accurate knowledge retrieval, and dynamic knowledge base updates to stay current with evolving regulations.
A ChatGPT-inspired UI was delivered in two weeks across both platforms. Infrastructure runs on Kubernetes/Hetzner, built to scale from hundreds to tens of thousands of concurrent users.
Brights built an autonomous AI consultant widget from scratch — no existing infrastructure, no off-the-shelf chatbot.
The process started with a PoC: a Telegram bot tested against 250 real expert questions in HR and tax law, scored by internal specialists. With 80% of answers rated excellent, the concept was validated and MVP development began.
The final solution features a multi-agent architecture for specialized query handling, RAG (Retrieval-Augmented Generation) for accurate knowledge retrieval, and dynamic knowledge base updates to stay current with evolving regulations.
A ChatGPT-inspired UI was delivered in two weeks across both platforms. Infrastructure runs on Kubernetes/Hetzner, built to scale from hundreds to tens of thousands of concurrent users.
Results
The widget deployed across both platforms in three months.
It handles thousands of simultaneous queries 24/7, achieving 80% expert-level accuracy with response times under 90 seconds.
Human consultants now focus on complex edge cases rather than routine questions.
The client expanded consultation capacity significantly without adding headcount, redirecting costs toward product development instead.
The widget deployed across both platforms in three months.
It handles thousands of simultaneous queries 24/7, achieving 80% expert-level accuracy with response times under 90 seconds.
Human consultants now focus on complex edge cases rather than routine questions.
The client expanded consultation capacity significantly without adding headcount, redirecting costs toward product development instead.
