Discover our blogs

The Hidden Foundation Behind Modern AI Systems

When people talk about AI, they usually focus on models, prompts, or tools.

But in real-world AI systems, something more fundamental is happening.

AI needs structure. AI needs memory. AI needs context.

In a recent session, Balaji Viswanathan highlighted a powerful idea:

Modern AI systems are not just built with code — they are built using three core formats

These formats act as the memory + communication layer between humans and AI agents.

Let’s break this down step by step.

Step 1: YAML (.yaml) – The Control Layer

Think of YAML as the brain configuration file of your AI system.

What makes YAML powerful?

✔ Human-readable
Unlike JSON, YAML avoids curly braces. It uses indentation (like Python), making it clean and easy to edit.

✔ Machine-parseable
Even though it looks simple, it is strict.
If the structure is wrong, systems immediately throw errors.

✔ Acts like a database
You can store structured information in a clean hierarchy.

Real Use Cases

• Defining prompt hierarchy
• Setting agent roles
• Managing permissions

Example:

• Base prompt (company-level)
• Feature-level prompt

• Role-based prompt (Coder, Reviewer, PM)

  Simple Mental Model

  YAML = Control Center

Diagram 1: AI Collaboration Layers

          					+----------------------+
          					|     YAML (.yaml)     |
          					|   Control Layer      |
          					| (Rules, Prompts,     |
          					|  Permissions)        |
          					+----------+-----------+
                     					|
          					+----------v-----------+
          					|   Markdown (.md)     |
          					|   Text Layer         |
          					| (Docs, Instructions, |
          					|  Communication)      |
          					+----------+-----------+
                     					|
          					+----------v-----------+
          					|     SVG (.svg)       |
          					|   Visual Layer       |
          					| (Diagrams, Charts,   |
          					|  Presentations)      |
          					+----------------------+

Step 2: Markdown (.md) – The Text Layer

If YAML is control, then Markdown is communication.

Why Markdown?

✔ Simple formatting
Use **bold**, *bullets*, # headings — no complex HTML needed.

✔ AI-friendly format
Most Large Language Models (LLMs) are trained heavily on Markdown.

✔ Universal usage
Used in:

• Documentation
• Notes
• AI instructions

Hybrid Format (Very Important)

• Modern tools combine YAML + Markdown.
• Used in tools like:
• Notion
• Obsidian
• Claude
• Structure:

							---
							title: AI Agent Instructions
							role: reviewer
							permissions: read-only
							---

							
							## Instructions
							Review the code and provide feedback...

YAML at top = metadata
Markdown below = actual content

Simple Mental Model

Markdown = Conversation Layer

Step 3: SVG (.svg) – The Visual Layer

Now comes the most underrated part — visuals.

Why SVG?

✔ Code-based graphics
SVG is not an image — it’s code.

✔ AI can generate it
Since it’s structured, AI can create diagrams easily.

✔ Highly precise
Perfect for:

• Architecture diagrams
• Charts
• UI mockups
• Slide decks

Real Use Cases

• Auto-generated architecture diagrams
• Competitive analysis charts
• PowerPoint slides

Simple Mental Model

SVG = Visualization Layer

Diagram 2: How AI Uses These Formats Together

									User Input
    									|
    									v
									+------------------+
									| Markdown (.md)   |
									| Instructions     |
									+------------------+
        									|
        									v
									+------------------+
									| YAML (.yaml)     |
									| Rules & Context  |
									+------------------+
        									|
        									v
									+------------------+
									| AI Processing    |
									| (LLM / Agent)    |
									+------------------+
        									|
        									v
									+------------------+
									| SVG Output       |
									| Visual Results   |
									+------------------+

Real Example: Competitive Analysis Sprint

To make this practical, let’s look at a real use case discussed in the session.

The Task

Build a competitive analysis tool

Example:
 A forecasting product using SAP data for CFOs

What AI Does

1. Reads Markdown instructions
2. Uses YAML context (industry, rules)
3. Generates SVG visualization

Output

1. A 2x2 matrix like:
2. X-axis → Organization Size
3. Y-axis → Maturity Level
4. Showing:
5. Competitors
6. Market positioning

Key Insight

Here’s the most powerful takeaway from Balaji Viswanathan:

“Communication to the AI is underestimated.”

If you know what to ask, AI becomes powerful
If not, even powerful AI becomes useless

Final Thoughts

Most people think AI success depends on:

❌ Models
❌ Tools
❌ APIs

But in reality:

✅ Structure (YAML)
✅ Clarity (Markdown)
✅ Visualization (SVG)

These three together define how effective your AI system will be.

Simple Summary

If you are building AI systems, start thinking beyond code.

Start thinking in formats, structure, and communication.

That’s where real AI engineering begins.