AI That Knows Your Business: How RAG Search Powers GTM OS Coaching

The problem with generic AI coaching

ChatGPT gives you advice for any business. It doesn't know your ICP. It doesn't know your pipeline. It doesn't know you've been stuck on the same objection for three weeks or that your positioning statement is still a rough draft. It starts from zero every time.

Your AI coach should give advice for YOUR business. The difference is context - and context is exactly what most AI tools strip out to stay general. We built the opposite. The coaching inside GTM OS pulls from everything you've put into the platform before it says a word.

How RAG works inside the OS

RAG stands for Retrieval-Augmented Generation. Before the AI responds to your question, it runs a similarity search across a vector database of everything relevant to you - your documents, your course lessons, your profile data.

The technical details: we use pgvector with OpenAI's text-embedding-3-small model, which generates 1536-dimensional embeddings for every chunk of content. When you ask a question, that question gets embedded into the same vector space. The database then runs a cosine distance search - chunks with a distance score below 0.3 are considered semantically close enough to be relevant and get passed to the AI as context.

Your uploaded documents and all 502 lessons in the curriculum get vectorized. The retrieval step happens before every coaching response, so the AI always has the most relevant material in front of it when it answers you.

What your AI coach actually knows

The RAG pipeline pulls from multiple sources tied to your account. At coaching time, the AI has access to:

When you ask "why am I losing deals at the demo stage," the AI doesn't answer generically. It cross-references your pipeline data, your DISC profile, and relevant demo course lessons before responding.

13 AI task types

The coaching layer handles more than free-form chat. There are 13 distinct task types built into the OS, each with its own prompt structure and context window:

AI coaching nudges

The daily nudge system is not generic motivation. It doesn't tell you to "keep going" or "stay focused." It notices patterns in your actual data and surfaces specific, actionable recommendations tied to what the numbers show.

If you haven't logged any outreach activity in three days, the nudge references that gap directly and links to the pipeline management lesson most relevant to your current stage. If your ICP score came back low on the specificity dimension, the nudge points you to Course 2 and tells you which exercise to prioritize. If your roleplay scores are consistently lower on discovery questions than on closing, the nudge flags that pattern and queues up the discovery simulation.

Each nudge is generated fresh from your current data - not templated. The AI reads your profile state, identifies the highest-leverage gap, and writes one specific recommendation. One thing. Not a list of ten suggestions that overwhelm and get ignored.

The book connection

The 98,000-word Founder and Small Team GTM is indexed in the same RAG pipeline as the course lessons. Every chapter, framework, and case study is chunked and vectorized.

When you ask your AI coach about the MAGNETS framework or DISC-based selling, it retrieves the relevant passages from the book alongside the matching course lessons and synthesizes them into a single response. You're not getting a generic summary - you're getting the specific section that applies to your question, alongside the lesson that teaches the skill.

The book and the platform are the same knowledge base. The RAG search doesn't distinguish between them. It finds what's most relevant regardless of where that content lives.

Support

The AI handles coaching, grading, roleplay, and content retrieval. But when you have a question about the platform itself, a feature you can't find, a billing issue, something that isn't working, that's not a job for AI.

For platform support, contact us at aistartuplaunch@gmail.com. Real human support, not a bot trying to deflect your question into an FAQ article. When you need a person, you get a person.

Frequently asked questions

What technology does the GTM OS use for its RAG search?

The platform uses pgvector with OpenAI's text-embedding-3-small model to generate 1536-dimensional embeddings for every chunk of content. When you ask a question, it runs a cosine distance search and pulls chunks with a distance score below 0.3 as context for the AI.

What specific data does the AI coach look at before answering my questions?

The AI pulls from your ICP document, positioning statement, DISC profile, pipeline stage, and course history. It also analyzes your assessment scores across 8 dimensions, uploaded pitch decks, proposals, and all 502 lessons in the curriculum.

What can the AI coach do besides answer open-ended questions?

The coaching layer handles 13 distinct task types, including roleplay conversations, assessment grading, discovery call prep, and email sequence generation. It also provides line-by-line artifact feedback on cold emails, objection handling responses, and ongoing readiness scoring.

How do the daily AI coaching nudges work?

The nudge system analyzes your actual data to surface one specific, actionable recommendation each day instead of generic motivation. For example, if you have not logged outreach in three days, it references that gap directly and links you to the most relevant pipeline management lesson.

How do I get support if I have a billing issue or platform bug?

The AI handles coaching, grading, and content retrieval, but platform support is handled by real humans. You can contact the team directly at aistartuplaunch@gmail.com to get help with billing, features, or technical issues.