KGT Solutions is an Industry 5.0 systems integrator operating across three practice areas: Industrial Automation (predictive maintenance, computer vision, IIoT, smart manufacturing, MES implementation), AI Solutions (custom LLMs, intelligent agents, data infrastructure), and Software & Platforms (SaaS products, AR/VR/XR, blockchain and Web3 systems). The company serves clients across manufacturing, automotive, steel, cement, chemical, pharma, oil & gas, energy, logistics, agriculture and healthcare - with live deployments including Tata Steel, Zuari Industries and Rapturous Solutions, from its offices in Noida and Ontario, Canada.
The trigger for this project came from the marketing team. Site traffic was rising, but none of it was converting into anything actionable: visitors browsed pages spanning a wide range of industries and service lines, found no interactive way to navigate the content for their specific situation, and left without leaving any trace. No name, no company, no industry, no problem statement - nothing for sales to work from. The marketing team proposed KAI as the fix: an AI navigation assistant with the full website indexed, able to answer service and industry-specific questions in real time, and structured to turn every conversation into a captured, qualified lead.
kgt.solutions carries detailed information about every service KGT offers - predictive maintenance, computer vision, MES integration, IIoT and more - but a visitor still had to know where to find it and what to ask. There was no way to surface the right content for a specific question, no way to understand what a visitor was actually trying to solve, and no way to collect their details before they left. Traffic was showing up. Leads were not.
The two gaps were discoverability and capture. A plant manager trying to find out whether KGT could handle a conveyor monitoring problem had no way to ask the site directly - they had to navigate manually across multiple pages, or send an email and wait. And even if they found what they were looking for, the only path to giving KGT their details was a generic contact form. The marketing team's diagnosis: the site needed a navigation layer with all of its content indexed that could answer questions in real time, while simultaneously collecting the three qualifying facts sales needed before a conversation was worth pursuing - industry, specific challenge, and budget range.
Before KAI, the number of visitors who left any contact detail at all was effectively zero - there was no mechanism to collect it short of a form submission. KAI changed that baseline immediately: from day one of deployment, the site was generating 2-4 qualified lead contacts per day, with roughly 10 people initiating a conversation with KAI daily.
For an early-stage company building its pipeline, that represents a complete shift from zero passive capture to a consistent, automated daily intake - with volume expected to grow as the team expands its reach.
KAI is a Claude-powered, retrieval-augmented assistant embedded directly into the KGT Solutions site, rather than a generic third-party widget. It runs as a FastAPI backend on a Dockerized VM: a ChromaDB vector store, populated from KGT's own service content via CPU-only sentence-transformer embeddings, retrieves the right context for each question, and the Claude API turns that into a qualifying conversation. Any visitor who shares enough detail gets pushed straight into the CRM as a structured, pre-qualified lead, and a webhook layer lets a salesperson step into that exact conversation the moment it's assigned to them.
Deployment breakdown:
Branded chat widget. Built and embedded a chat widget (KAI) matching KGT's visual identity - teal accents, a dark header matched to the site's navbar, a rounded launcher button, a welcome message with four quick-reply chips, a typing indicator, and persistent conversation state.
Scoped system prompt. Wrote a system prompt that scopes KAI strictly to KGT's actual service lines (Industrial AI & Automation, Computer Vision, IIoT, Predictive Maintenance, MES/Industry 4.0-5.0, Web3/XR/Blockchain) and drives every conversation toward three qualifying facts - industry, specific challenge, budget range - before suggesting a consultation booking.
RAG content index. Indexed KGT's own service and case-study content into a ChromaDB vector store using CPU-only sentence-transformer embeddings, giving KAI retrieval-augmented, on-brand answers instead of generic model output.
CRM integration. Wired the site's enquiry and schedule forms directly into the CRM - first HubSpot, then migrated to Zoho - writing six custom lead fields (industry, primary pain point, budget range, lead source detail, acquisition page, session ID) so every chatbot-sourced lead lands pre-tagged for sales.
Human takeover system. Built a webhook-driven takeover layer (/takeover, /release, /agent-reply), triggered by CRM workflow rules the moment a lead is assigned, so a human rep can drop into a live KAI conversation without the visitor repeating themselves.
Tech Stack:
Python · FastAPI · Docker
ChromaDB · sentence-transformers (CPU-only) · Anthropic Claude API
BeautifulSoup4 / Requests for content ingestion
PostgreSQL via Prisma / asyncpg
Zoho CRM (migrated from HubSpot) via OAuth 2.0 with CRM-side workflow webhooks
Timeline: 6 weeks, early May to mid-June 2026. The initial widget - branded UI, RAG pipeline and Claude API integration - was built and live on the site first. The CRM integration (HubSpot, then the full Zoho migration including the takeover webhook system) was completed across the remaining weeks, with the full end-to-end system operational by mid-June.
KAI has been live on kgt.solutions since May 2026. The numbers below reflect current daily performance; as KGT scales its outreach, both conversation volume and lead capture are expected to grow proportionally.
Metric | Before KAI | After Deployment |
|---|---|---|
Daily website visitors who engage and provide contact details | 0 - no capture mechanism existed | 2-4 qualified leads captured per day |
Daily chatbot conversations initiated by visitors | 0 | ~10 conversations/day and growing |
Lead data arriving pre-tagged in CRM (industry, pain point, budget) | 0% - all fields collected manually after first contact | 100% of KAI-sourced leads arrive pre-tagged |
Visitor path to asking a service question | Manual navigation across static pages or email with 24- hr+ wait | Answered in real time via KAI, 24/7 |
The structural shift goes beyond the headline numbers. Every lead KAI captures arrives in Zoho pre-tagged with industry, primary pain point and budget range - data sales previously had to chase down over email or on a first call. A rep picking up a KAI-sourced lead already knows what sector the person is in, what problem they described, and roughly what budget they are working with before saying a word. The takeover webhook layer compounds this: when a lead gets assigned, the rep can step into the live conversation directly rather than starting cold with a “thanks for reaching out” email.