README.md

---
title: Agent Ken — Data-Informed PM Copilot
emoji: 🤖
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: "6.5.1"
app_file: app.py
pinned: true
license: mit
short_description: Data-informed PM Copilot
---

# 🤖 Agent Ken — Your AI Companion for Product Management & Data
An AI assistant that helps you **learn and practice product management** — from strategy, frameworks, and execution to real-world data analysis.
Agent Ken is also integrated by **Microsoft Fabric** with real product analytics: **5,000 users**, **204K events**, **90 days of metrics**, and **3 trained ML models** (churn, LTV, anomaly detection). But it's not just about data — you can also ask about **PM frameworks, AI & tech, business strategy, and more**.
> **This isn't a generic ChatGPT wrapper.** Ask about retention, and you'll get actual numbers. Ask about PRDs, and you'll get a structured template. Ask about AI, and you'll get practical insights.

## ⚡ Powered By

| Component | Technology |
| :--- | :--- |
| 🧠 **AI Agent** | Azure AI Foundry Agent Service (GPT-5) |
| 🏭 **Data Platform** | Microsoft Fabric (OneLake + MLflow) |
| 🤖 **ML Models** | Churn Prediction · LTV Estimation · Anomaly Detection |
| 🖥️ **Frontend** | Gradio (Hugging Face Spaces) |

## 🚀 What Can Agent Ken Help You With?

### 📊 Data-Informed (Powered by Fabric ML Pipeline)
Ask about real product metrics — Agent Ken answers with **specific numbers**, not vague advice:

* "What's our D7 retention?" → **25.6%**, with segment breakdown by channel.
* "Which channel has highest LTV?" → **Organic at $25.65** (1.9x vs Paid Social).
* "Any anomalies in our metrics?" → **Nov 24-25 churn spike** (2.8x average).
* "How did checkout experiment do?" → **+11.9% conversion**, p=0.027, ship it.

### 🧠 General PM (Powered by GPT-5.1 Knowledge)
All the PM capabilities you'd expect from a senior product manager:

* 🔍 **Product Discovery** — Jobs-to-be-Done, personas, problem statements, user journeys.
* 📊 **Prioritization** — RICE scoring, MoSCoW, trade-off analysis.
* 📝 **PRDs & User Stories** — Product requirements, acceptance criteria, MVP definition.
* 🧪 **Experiment Design** — A/B tests, hypothesis templates, success metrics.
* 📈 **Metrics & OKRs** — North Star Metrics, funnel metrics, guardrail metrics.
* 🤝 **Stakeholder Communication** — Decision memos, release notes, progress updates.
* ♿ **Inclusive Design** — Accessibility, ethical considerations, vulnerable user safety.

---

## 💾 What Data Powers Agent Ken?

A synthetic but realistic product analytics dataset simulating a mobile app over 3 months:

| Dataset | Records | What's Inside |
| :--- | :--- | :--- |
| **Users** | 5,000 | Channel, country, plan, age group |
| **Events** | 204,356 | App open, purchase, support ticket, etc. |
| **Daily Metrics** | 90 days | DAU, revenue, retention, churn, NPS |
| **A/B Tests** | 5 experiments | Checkout, onboarding, push timing, pricing, dark mode |

### 🔑 Key Numbers

| Metric | Value |
| :--- | :--- |
| Avg DAU | **980** |
| D7 Retention | **25.6%** |
| Avg LTV | **$19.87** |
| Churn Rate | **34.2%** |
| 90-day Revenue | **$99,241** |

---

## 🤖 ML Models Trained in Fabric

| Model | Algorithm | Score | Key Insight |
| :--- | :--- | :--- | :--- |
| **Churn Prediction** | Gradient Boosting | AUC: 1.00 | Users inactive 3+ days = high risk |
| **LTV Estimation** | Linear Regression | R²: 0.77 | Organic users worth 1.9x more than paid |
| **Anomaly Detection** | Isolation Forest | 5 anomalies | Nov 24-25 churn spike = possible incident |

## 🧪 A/B Test Results

| Experiment | Lift | Significant? | Action |
| :--- | :--- | :--- | :--- |
| **Checkout Redesign** | +11.9% | ✅ p=0.027 | **Ship** ✅ |
| **Onboarding Simplification** | +8.0% | ✅ p=0.021 | **Ship** ✅ |
| **Push Notification Timing** | +22.0% | ✅ p=0.035 | **Ship** ✅ |
| **Premium Pricing Test** | -5.0% | ❌ p=0.289 | **Rollback** ❌ |
| **Dark Mode Default** | +3.0% | ❌ p=0.212 | **Extend test** ⏳ |

---

## 💡 How to Use

1.  **Type your question** in the chat box.
2.  **Upload files** — drop your CSV, PDF, Excel, or other files. I'll analyze them and give you actionable insights.
3.  **Ask about data** — Agent Ken will cite specific numbers from the Fabric pipeline.
4.  **Ask about PM** — Agent Ken will use product management expertise.
5.  **Try the examples** — click "Try an example" for inspiration.
6.  **Start fresh** anytime with "New Conversation".

## 🎯 Tips for Best Results

* ✅ *"What's our retention by acquisition channel?"*
* ✅ *"Help me write a PRD for a referral program based on our LTV data"*
* ✅ *"Score these features using RICE: push notifications, dark mode, onboarding revamp"*
* ✅ *"Design an A/B test for our new premium pricing"*
* ❌ Off-topic questions (biology, cooking, etc.) — Agent Ken is scoped to PM & tech.

## 👤 About

Built by **Kendrick Filbert** — AI + PM + Social Impact Practitioner

*Powered by **Azure AI Foundry** (GPT-5.1) · **Microsoft Fabric** (OneLake + MLflow) · **3 ML Models** (Churn · LTV · Anomaly)*