Veridian

Project Overview

Veridian is a B2B SaaS whose sales team was drowning in demo requests — most of them unqualified. We built an AI assistant that pre-screens every visitor before a single rep picks up the phone: it answers product questions from the actual documentation, captures intent and contact details, and only forwards qualified requests to sales. The result is a pipeline that filters itself, 24/7, in three languages.

Client Challenges

• The sales team spent more than half their day answering questions that were already answered in the documentation (pricing, features, onboarding).
• Answers were inconsistent between support and sales, which eroded prospect trust at the exact moment of evaluation.
• Veridian had zero visibility in AI search engines — prospects researching on ChatGPT, Perplexity or Gemini saw competitors instead, even where Veridian was objectively the better fit.

Our Solution

• A RAG chatbot powered by Gemini 2.5 Flash Lite, connected to Veridian’s documentation through a pgvector vector store. Every answer is grounded in a real source, always up to date, and never contradicts the docs.
• A complete admin panel to manage bots, sources and conversations — the team can add a source or tune a bot without touching any code.
• GEO applied to Veridian’s content and documentation so the brand gets cited in generative-engine answers, not just in Google.

Key Features

• Grounded AI assistant: answers exclusively from the indexed documentation, with a visible source reference on every reply.
• Lead pre-qualification: firmographic + intent detection before handoff — sales only sees qualified demos, with context attached.
• Bot management admin: create, configure and monitor bots, sources and conversations from one panel.
• Embeddable chat widget: a few lines of JavaScript drop the assistant anywhere on the site.
• Conversation analytics: every interaction is tracked to spot content gaps and coach the team.

Tech Stack

• LLM: Gemini 2.5 Flash Lite
• Retrieval: RAG + pgvector (vector database)
• Admin panel: Filament (Laravel / PHP)
• Widget: lightweight embedded JavaScript chat

Results & Impact

• 78% of demo requests are now handled end-to-end by the assistant, with no human involved.
• +41% qualified demos booked after pre-screening was introduced.
• 63% less time spent by sales on unqualified or repeat questions.
• 96% answer consistency across support, sales and the chatbot.
• 87% of AI-generated comparisons now cite Veridian as a source on ChatGPT, Perplexity and Gemini.

“Sales now talks to the right people, and the assistant never contradicts our pricing. It feels like a second team that never sleeps.” — Head of Revenue, Veridian