I Built an AI That Listens to Me Live and Suggests Links 🤯 | Hero Extract

*Summary*
In this Hero Club extract, the host walks through a real-time AI-powered tool he built in just two days. The system listens to live conversations, matches spoken topics against a website content database using semantic embeddings, and automatically suggests or publishes relevant links to an OBS banner, Telegram, or other outputs. The discussion covers the full architecture from creating a custom WordPress API plugin to embedding-based concept matching.

*Key Points*
[00:00:00] - Building a Custom API and the Initial Idea
• The conversation started after Ryan presented his Cursor workflow, which sparked the idea of having AI watch live conversations and auto-suggest relevant content.
• Instead of using an existing MySQL connection to the Automator site (which is complicated to set up for others), the host decided to build a custom WordPress API plugin using AI.
• AI generated the plugin, which consolidates pretty links, downloads, pages, and posts into a single API endpoint - making the data easily accessible and shareable with other WordPress users.
• This was the host's first time creating an API despite having used hundreds of them, highlighting how AI lowers the barrier to building new tools.

[00:02:13] - Embeddings, Real-Time Transcription, and Concept Matching
• The host switched from Handy to Assembly AI for real-time transcription because Handy ran too slowly on his laptop.
• Website data is pulled as JSON from the API, converted into embeddings, and stored in a local SQLite database.
• A key insight: the system matches on concept rather than exact words, which dramatically improves the quality of results.
• Assembly AI watches everything being said in real time, embeds the spoken text, and compares it against the stored content database for matches.
• A threshold system controls output: high-confidence matches auto-publish, medium matches are shown for manual approval, and low matches are discarded entirely.
• The entire working prototype was built in roughly two days.

[00:04:25] - Live Demo, Output Options, and Future Plans
• The tool updates an OBS banner in real time with matched content links, replacing the manual process of searching for and posting relevant links during live calls.
• Future plans include pushing matched links to Telegram for persistence (since live chat messages disappear) and potentially to YouTube live stream chats.
• The notification display and transcript chunk size are configurable, and the host can route the display to a private monitor so viewers do not see the suggestion interface.
• The group reacted enthusiastically, recognizing this as a near cutting-edge application of real-time AI during live conversations.
• The host emphasized that the most impressive aspect was using embeddings for concept-based matching rather than simple keyword matching.

*Brief Summary*
This clip showcases a practical, real-time AI assistant that listens to live conversations, semantically matches topics against a website content database, and surfaces relevant links automatically. Built in just two days using a custom WordPress API plugin, Assembly AI transcription, Python, and embedding-based matching in SQLite, it demonstrates how accessible these advanced AI workflows have become for creators and community leaders.

#AIEmbeddings #Real-TimeTranscription #AssemblyAI #CursorTool #WordPressAPI #ConceptMatching #SQLiteDatabase #PythonAutomation #OBSStreaming #TelegramIntegration #HKHeroClub #LiveContentSuggestions #CustomPlugin #SemanticSearch #AIAutomation

Видео I Built an AI That Listens to Me Live and Suggests Links 🤯 | Hero Extract канала AUTOHOTKEY Gurus