3-Tier Knowledge System

Structured knowledge injection powering every TorlyAI skill

What It Does

Every AI agent in TorlyAI is backed by a structured knowledge pipeline built from 77,000+ words of UK Innovator Founder Visa guidance. Instead of dumping all that context at once, the system loads only what each skill needs — keeping responses fast, focused, and accurate.

The system compresses source material from ~2.4 million tokens down to 4,000–8,000 tokens per skill invocation — roughly 2–4% of the original, with no loss of actionable content.

The Three Tiers

1

Tier 1 — Core Knowledge

Always injected into every agent's system prompt. Contains the fundamental frameworks that every skill needs:

  • The Impossible Triangle (Innovation Noose, Viability Guillotine, Scalability Trap)
  • 4F Formula: PMF (30%) + FMF (25%) + BMF (25%) + Fortune (20%)
  • Business Plan section definitions (7 sections)
  • Endorsement body summaries (UKES, Innovator International, Envestors)
  • Score thresholds: Excellent (80+), Good (65–79), Fair (50–64), Poor (<50)

~2–3K tokens

2

Tier 2 — Skill Knowledge

Per-skill modules loaded only when that skill runs. Each skill has a dedicated knowledge file with domain-specific expertise.

Innovation Matrix — 4F sub-dimensions, Canvas blocks
Financial Modeler — 7 financial cognitions, I-P-O-C
Endorsement Navigator — rejection reasons, body selection
Interview Coach — 3 verification goals, 8 common mistakes
Caseworker Checker — UKES evaluation points, TRL scale
Documents Checklist — full document requirements

~1–5K tokens per skill

3

Tier 3 — Reference Material

On-demand chunks extracted by heading from five learning modules. Loaded only when an agent needs deep detail on a specific topic.

ModuleTopic
Module 1Innovation Matrix (IM)
Module 2Business Plan (BP)
Module 3Financial Modeling (FM)
Module 4Endorsement Process (EP)
Module 5Interviews & Pitch Decks (IP)

Variable — loaded by heading on demand

How It Works

When you invoke a skill, the knowledge loader follows this sequence:

  1. 1
    Core knowledge is always present — the 4F Formula, Impossible Triangle, and endorsement body context are already in the agent's system prompt.
  2. 2
    Skill knowledge is injected — the loader finds the matching skill module and adds its specialised content to the prompt.
  3. 3
    Reference chunks are loaded on demand — if the agent needs deep detail on a specific topic, it requests the relevant heading from the matching module.

Why It Matters

Accuracy

Every response is grounded in official visa guidance, not generic AI knowledge.

Efficiency

Only relevant knowledge is loaded, keeping token usage low and responses fast.

Consistency

All agents share the same core framework, so advice never contradicts across skills.

Extensibility

New skills get knowledge injection automatically by adding a knowledge module.

Related

Innovatorly Matrix

The 4F Formula that forms Tier 1 core knowledge

All Skills

See the skills that use this knowledge system