Scaliam Multi-Tenant RAG Support Platform
Scaliam is an AI support platform we built, launched and operate ourselves. A business points it at their website or uploads their own documents, Scaliam crawls that content, converts it into a searchable vector database, and installs a chat widget that answers visitors' questions from their own material rather than the open internet or guesswork. When a question falls outside what it can answer confidently, the conversation is handed to a human instead of being answered anyway.
Project Overview
Scaliam is an AI support platform built, launched and operated by our own team. It crawls a client's website or accepts uploaded documents, converts that content into a searchable vector database, and installs a chat widget that answers visitors' questions from the client's own material. It runs at a 95% answer success rate, serves multiple client workspaces from one multi-tenant system, and is the product our team maintains, bills, monitors and answers for daily.
Core Features
- Automatic content ingestion, crawling a client's site or accepting uploaded data and keeping the knowledge base in sync.
- Vector search on Supabase with pgvector, retrieving answers by meaning rather than keyword matching.
- Multi-tenant workspaces, with each client's content, settings and conversations fully isolated.
- Embeddable widget installed with one script tag, styled to match the host site rather than looking bolted on.
- Engineered human handover, where a confidence threshold decides when not to answer and routes the visitor to a person.
- Grounded answers only, constrained to the client's own content with no invented facts.
- Conversation history and insights showing what visitors actually ask.
Technical Highlights
- Multi-tenant architecture isolating each client's content, settings and conversations within one shared system.
- Semantic retrieval built on Supabase (PostgreSQL and pgvector) rather than keyword search.
- Confidence-based handover logic that hands a conversation to a human when the system cannot answer reliably.
- Answers constrained to a client's own ingested content, avoiding invented or ungrounded responses.
- Built on Next.js and Node.js, with OpenAI powering the answer generation layer.
Project Impact
Scaliam runs in production at a 95% answer success rate, serving multiple client workspaces from a single multi-tenant system that our own team operates day to day. By grounding every answer in a client's own content and handing off to a human when confidence drops, it gives businesses an AI support layer that stays accurate instead of guessing, while also surfacing what visitors are actually asking.
Technologies Used
Project Details
- Custom AI Development
- Agentic AI Development
- SaaS Platform Architecture
