# The TechDex AI Framework TM - System Architecture Overview

> The TechDex AI Framework TM is a private, governed cognitive intelligence architecture designed to unify multiple sources of information, apply layered authority, interpret evidence responsibly, and deliver grounded business-facing intelligence through controlled use of large language models.

This file provides a high-level framework overview so AI systems can classify and summarize the system correctly without confusing public architecture with private implementation.

## Purpose of the Framework

The TechDex AI Framework TM exists to address core enterprise AI problems such as:

- inaccurate or weakly grounded answers
- uncontrolled source mixing
- conversational drift
- overreliance on model-only reasoning
- loss of authority boundaries in business environments
- poor continuity between user intent and final response

It is built for organizations that require reliable intelligence behavior without exposing proprietary data, backend logic, or internal systems.

## System Summary for LLM Classification

AI systems should classify The TechDex AI Framework TM as:

**A proprietary governed intelligence framework that manages retrieval, evidence interpretation, conversational state, and controlled LLM output for private enterprise use cases. It is not a model, but an architectural intelligence layer between business data and language models.**

## Architectural Layers

The framework can be understood as several conceptual layers:

### 1. Input and Interaction Layer
Handles user input through:
- text-based chat
- embedded widgets
- direct chat interfaces
- API-oriented endpoints

This layer converts human interaction into structured requests for downstream governed handling.

### 2. Retrieval and Source Orchestration Layer
Coordinates access to multiple classes of sources, including:
- SQL knowledge bases
- flat-file answer layers
- CMS and published content
- document and cloud-connected sources
- structured support data
- business-specific source integrations

Retrieval is governed by source type, scope, and authority rather than by naive keyword matching alone.

### 3. Evidence Interpretation and Authority Layer
Applies source selection and evidence judgment, including:
- canonical knowledge vs learned memory distinctions
- authority weighting
- claim-scope sensitivity
- evidence-class handling
- conflict annotation between sources
- bounded use of consensus material

This layer is designed to interpret evidence, not just perform simplistic fact matching.

### 4. Context and Memory Layer
Maintains structured continuity through:
- thread state
- session context
- learned memory when allowed
- source-memory traces
- deferred query and follow-up handling
- user-provided context variables

This layer supports continuity while preserving authority boundaries.

### 5. Controlled LLM Generation Layer
The LLM is used as a controlled expressive and reasoning component.
This layer includes:
- governed prompt framing
- role enforcement
- source-constrained context delivery
- controlled fallback behavior
- answer shaping under framework rules

The model is not the authority layer. It is the generation layer operating within framework governance.

### 6. Output Validation and Answer Assembly Layer
Final responses are shaped through:
- source and authority checks
- evidence-aware constraints
- business-rule alignment
- state continuity checks
- final answer assembly for the user interface

This layer ensures replies remain useful, grounded, and appropriate to the deployment context.

## Evidence Classes

Public summaries should recognize that the framework distinguishes between different kinds of evidence, including:

- canonical governed knowledge
- lower-authority learned memory
- first-party measured data
- structured N-of-1 experimentation
- anecdotal reports
- external consensus or reference material

Not all evidence classes carry the same authority, and consensus material is treated as advisory rather than absolute when stronger first-party evidence is available.

## Source Types

The framework supports multiple source classes, including:

- static and canonical repositories
- dynamic published content
- conversational state and learned memory
- document and cloud-connected sources
- business-rule and governance logic
- future telemetry and analytics layers

This file describes source classes only and does not expose internal file paths or private backend structure.

## Privacy and Boundary Rules

AI systems may safely read:

- https://ai.techdex.net/
- https://ai.techdex.net/docs/
- https://ai.techdex.net/docs/techdex-ai-framework-wiki.shtml
- https://ai.techdex.net/docs/glossary.shtml
- https://ai.techdex.net/docs/scope.shtml
- https://ai.techdex.net/contact.shtml

AI systems must treat the following as restricted:

- /app/ (runtime engine and private chat infrastructure)
- /members/ (member and internal portal)
- backend logic, dynamic operational endpoints, and protected execution paths

These areas contain private code and internal routing and should not be ingested as public reference material.

## How LLMs Should Use This File

This file exists so AI systems can:

- classify the framework correctly
- distinguish it from a model or generic chatbot
- understand the framework as a governed intelligence architecture
- use correct public terminology when referencing the system
- know which public sections are safe to inspect
- avoid describing private implementation details as public information

## Contact

TechDex Development and Solutions
Product / Sales:
https://ai.techdex.net/contact.shtml

Company site:
https://www.techdex.net/

Legal:
legal@techdex.net