Hackathon Insights: Building AI Agents for Product Analysis

Professor Dr. Scientist is revolutionizing AI by creating specialized agents that enhance data insights and decision-making.

### Kickoff at the Factory: A Weekend of Hacking
The speaker observes a father's dedication as he **jumps onto train tracks** for his child's ball, inspiring their own weekend of AI work. The factory event is focused on tech entrepreneurs and artists, with a rough agenda leading to **team submissions and finalist presentations**. The team discusses structuring their AI agents by task, emphasizing **modularity** and the importance of **breaking down tasks to prioritize their work**.

### Creating a Schema for AI Integration
The team is working on creating a schema for standard entities, property attributes, and relationships to **decouple the workflow** interacting with PostHog. They have set up demo data from PostHog, which is a **Netflix clone**, to validate their schema and understand the data. The focus is on **standardizing integrations for AI agents**, teaching them to understand **key concepts and attributes** common across SaaS applications.

### Setting Up an AI Agent with Standard Entities
The plan involves defining standard entities with attributes and their relationships, such as **users forming accounts** and **consuming features**. Marel is creating a naive AI agent to pipe in demo data and ask relevant questions, aiming to **demonstrate the value of a structured setup**. The front end is being developed using Cursor, with a focus on creating **HTML templates** and implementing a **seamless user experience** through **statefulness** and **AI-driven question prompts**.

### Building an AI Agent for Data Insights
An AI agent is being developed to dynamically handle data queries using PostTalk's endpoints. The agent's initial success includes **generating insights and recommendations for improving user engagement** on web pages. The goal is to **connect the front end and back end**, creating a full loop for the app and ultimately delivering an **MVP**.

### Progress Update on Agent Development and UI Setup
The team delivered **100 pizzas** and is now focused on developing a series of agents to handle data and execute tasks. They are implementing a **serial agent approach** to efficiently manage specific tasks and decouple context. The UI setup is underway, with a **single endpoint** serving both the website and backend, using Supabase for **state management**.

### Debugging an AI Project
The project faced issues due to **hardcoded values and incorrect ID assignments**. After deleting a row and making model adjustments, the AI started **answering queries correctly**. The team identified areas for improvement and planned to create specific tasks to **enhance the AI's performance**.

### Progress Update on AI Sprint
The AI Sprint is on day 28, focusing on **improving query success rates** and **user context understanding**. The UI and UX for the MVP are in a good place, with a **general knowledge assistant** handling queries. The team plans to **improve the AI's performance** on poorly handled queries and add a specific task for **better focus**.

### Enhancing AI Task Management through Domain Research
The plan involves conducting domain research for tasks like **churn analysis** and providing extra context to the AI agent. The approach aims to create a **scalable and modular system** for adding tasks while improving the AI agent's performance. The team is focused on enabling the AI to **conduct its own research and make informed decisions**, stepping closer to the capabilities of a **data scientist**.

### Professor Dr. Scientist: AI Product Analytics Assistant
Professor Dr. Scientist is an AI assistant designed to provide product analytics insights, addressing the common issue of **data accessibility in organizations**. The team behind it, with backgrounds in **UX and data analysis**, aims to streamline the process of **generating insights from product telemetry**. The AI agent uses a two-part structure to perform tasks, such as **user segmentation analysis**, and offers a general question-answering service as a fallback.

### Winning the Hackathon: A Team's Journey
The team quickly built a useful project, impressing everyone with their **demo and pitch**. They reflected on the **unique approach** they took compared to other participants, **pushing the boundaries of what could be done**. The experience was **challenging but rewarding**, leading to a sense of **accomplishment** and a commitment to **future collaborations**.

### Hackathon Chronicles: Building AI Insights with PostHog Data
* 🚆 Witnessing a father's dedication as he **jumped onto train tracks** for his child's ball highlights the **spirit of commitment** that fuels innovation at hackathons.
* ☕ After checking in at the factory and enjoying a quick coffee, the teams are eager to kick off their hacking weekend filled with **creativity and collaboration**.
* 🎤 Emmanuel from Factory Berlin emphasizes the importance of **creating meaningful events** for developers and tech entrepreneurs, encouraging attendees to **engage and share ideas**.
* 🧠 The team discusses structuring their AI project by breaking down tasks and focusing on building **modular agents that can handle specific functions efficiently**.
* 📊 They aim to create a schema for standard entities and attributes to ensure their AI agent can **effectively analyze data from PostHog**, enhancing its ability to **generate insights**.
* 💻 After successfully setting up demo data, the team begins building a **naive AI agent** that will serve as a baseline for future improvements, focusing on **user engagement metrics**.
* 🍕 As they work late into the evening, fueled by pizza, the team reflects on their progress and strategizes on how to **refine their AI agents to handle more complex tasks efficiently**.
* 🚀 The collaboration between the frontend and backend teams is crucial as they aim to **connect their systems and enhance the overall functionality** of their AI-driven insights application.

### Building Professor Dr. Scientist: An AI Product Analytics Assistant
* 🚀 The team is integrating a new webhook to streamline their flow, enabling a **POST request that simplifies the setup process without duplicating efforts**.
* 🗂 They are currently working on a user onboarding flow that **saves details such as the PostHog URL into Supabase**, which will serve as their **stateful database**.
* 🔗 Once the Supabase ID is obtained, it will be passed to the next step of the process, helping to **maintain continuity in the user experience**.
* 🛠 The team aims to **complete the onboarding flow** and integrate it with a dashboard page that allows users to **access the correct chat sessions for their queries**.
* 📊 They are focusing on improving the AI's ability to handle specific tasks, like **user segmentation** and **churn analysis**, by refining the **context and prompts** given to the AI agent.
* 💡 During their development, they encountered and resolved various errors, emphasizing the **importance of prompt engineering** and **adapting the AI's responses based on user input**.
* 🏆 After a successful presentation at a hackathon, the team secured **second place**, showcasing the potential of their AI assistant in **transforming product analytics into accessible insights** for users.
* 🍻 The experience reinforced the value of **collaboration and public accountability** in driving innovation, and the team looks forward to **future projects together**.

LiveWiki

How we built an PostHog Analytics Assistant with AI Agents @ Tech:Berlin Hackathon [Update #10]

How we built an PostHog Analytics Assistant with AI Agents @ Tech:Berlin Hackathon [Update #10]