Project Description

I built a lightweight “conversation compass” that helps small businesses without huge marketing budgets instantly find high-value discussions happening online about their space. Instead of manually searching across platforms, the system scans the web in real time, ranks the best engagement opportunities, and generates copy-paste-ready replies.

How It Works

CopilotKit (Agent Orchestration)

CopilotKit drives the entire workflow. The agent takes a simple business description and then triggers tool calls to:

• Search for relevant conversations
• Load and store state in Redis
• Rank results

Everything is done dynamically through tool calls rather than fixed prompts.

Tavily (Search + Extraction)

Tavily provides live discovery of conversations on Reddit, X, and across the web. It returns titles, snippets, URLs, and extracted content the agent uses to classify intent and determine whether a thread is worth engaging.

Redis (Memory + Personalization)

Redis gives the system a stateful memory layer:

• Deduping threads so the user never sees the same one twice
• Saving timestamps and run history
• Storing thumbs-up/thumbs-down feedback to boost or suppress similar conversations in future runs

User Experience

The user describes their business, and the agent returns a curated list of conversations with:

• Platform labels
• Intent tags
• “Why this matters” context
• One-click Generate Response buttons
• Feedback controls tied into Redis

Why It’s Useful

For solo founders and small teams, this turns community engagement into a focused, automated workflow. It surfaces the right conversations at the right time, learns from feedback, and makes it effortless to jump in with helpful, relevant responses.