Back to portfolio
RAG / Knowledge · Customer Success

Enterprise Knowledge Base Chatbot

Enterprise-Website Knowledge Chatbot

Users need accurate answers about an enterprise professional-services firm's offerings → the AI retrieves and explains only the firm's website content → users get consistent, on-brand guidance without external noise.

01The Problem

Large professional services websites contain extensive, fragmented information across services, industries, and insights. Users struggle to find precise, trustworthy answers quickly, especially when navigating complex offerings like audit, tax, advisory, and consulting. Without a controlled interface, responses risk being inconsistent, outdated, or misaligned with the firm's official messaging.

02What the AI Does

* Retrieves and synthesizes information strictly from a provided knowledge base derived from the firm's website content * Summarizes, explains, and structures content about services, industries, and insights * Answers user questions using constrained retrieval (RAG-style) rather than general knowledge * Refuses to answer when information is not present in the source content * Guides users to relevant sections (e.g., audit, tax, consulting, industries) within the firm's site Built on: * GPT-5.3 (ChatGPT) * File-based knowledge retrieval (llms.txt as the sole source of truth) * No external APIs, browsing, or tool augmentation beyond the embedded document Configured to: * Restrict outputs to a single authoritative corpus (the firm's website content) * Maintain a professional, brand-aligned tone * Avoid speculation or general industry knowledge * Ask clarifying questions when queries are ambiguous

03Design Decisions

01 · Choice

Single-source knowledge constraint (firm website only)

Why

Ensure accuracy and brand alignment in all responses; avoid hallucinations or conflicting external information

Constraint

(firm website only)

02 · Choice

Use of llms.txt as embedded knowledge base

Why

Centralizes website content into a structured, machine-readable format for reliable retrieval [Creator: add rationale]

Constraint

Limits the assistant to static, pre-curated content rather than live updates

03 · Choice

Explicit prohibition of general knowledge or external sources

Why

Prevents dilution of official messaging and reduces legal/reputational risk in professional services context

Constraint

The AI may appear less helpful when users ask broader or comparative questions

04 · Choice

Professional, corporate tone calibration

Why

Aligns with the firm's positioning as a global audit, tax, advisory, and consulting practice

Constraint

Responses avoid casual or creative phrasing, prioritizing clarity and credibility

05 · Choice

Mandatory uncertainty handling (explicit “not available” responses)

Why

Ensures transparency and avoids fabricated answers in a high-trust domain

Constraint

Can result in more “no answer” responses compared to general-purpose assistants

06 · Choice

Clarification-first behavior for vague queries

Why

Website content is broad (services, industries, insights), requiring disambiguation to retrieve relevant sections

Constraint

Adds friction for users expecting immediate answers

07 · Choice

No tool use beyond static knowledge file

Why

Maintain tight control over data sources and eliminate variability from external integrations [Creator: add rationale]

Constraint

No real-time data, personalization, or dynamic lookup capability

05Key Insight

Constrained AI systems deliver higher trust and brand consistency by narrowing scope—even at the cost of breadth and flexibility.