SessionIQ — AI Session Intelligence for Coaches

🧠 The Problem Coaches don't usually lose clients over one bad session. They lose them in the follow-ups that never happened, or the warning signs nobody caught in time. SessionIQ was built to close that gap. ⚙️ How It Works A two-stage automation pipeline on n8n, powered by Groq's Llama 3.3 70B. Stage 1 — Takes raw, unstructured session notes and extracts action items, risks, decisions, and sentiment, enforcing a strict JSON schema so the output is consistent every time — not just usually right. Stage 2 — Cross-references that data against the client's session history to generate a 1–10 risk score and specific coach recommendations. 🟢 When Things Are Going Well If sentiment is positive and the risk score stays low, the system stays quiet. The coach's report shows a "healthy" status, the recap email goes out with a normal tone, and everything logs and moves on — no noise, no unnecessary alerts. 🔴 When a Client Starts Slipping The moment the AI detects negative sentiment, missed action items, or concerning language in the notes, the risk score climbs. At a score of 7 or higher, a high-risk flag triggers automatically — the coach's report changes tone, surfaces the specific warning signs it caught, and pushes direct recommendations instead of a routine summary. That's the difference between a coach finding out a client is disengaging three weeks later, and finding out the same day. 📤 What Happens End-to-End Once a session is logged, everything runs with zero manual steps: a branded recap email to the client, a confidential report to the coach, follow-up tasks in Google Tasks, and calendar holds for next steps. 🔧 The Hard Part Wiring the APIs together was the easy half. The real work was making the AI's output reliable enough to trust without a human checking it every time — schema enforcement, duplicate detection, and error handling took as long as the core pipeline logic itself.

🎥 Full architecture walkthrough: youtu.be/SGbv75lRRqg