A LITTLE CONTEXT
A chatbot is easy to put on a website. Giving it useful answers, connecting it to the business, and making sure a person can take over is a much more interesting problem.
Hello Marky Chat is my exploration of that whole experience. Its chat product helps a business answer visitors using its own knowledge, collect the context it needs, and carry the conversation into a shared inbox when the team needs to step in.
The answer starts with the business.
An assistant should understand the material a team gives it. Hello Marky Chat brings knowledge management and retrieval into the product, so the conversation can draw on business information instead of relying on a generic answer.
That work includes the less visible parts of an AI application: preparing source material, finding relevant context, managing conversation history, and deciding which tools the assistant can use. The widget is the front door to a larger system.
Keep the thread when a person takes over.
Some visitors need a quick answer. Others need a real conversation. The shared inbox lets the team see what has already been said and continue from there, rather than making the visitor explain the same thing twice.
Lead capture belongs in that flow too. The goal is to collect useful contact details and intent while the conversation is happening, then make that information available to the people who will follow up.
One product, several ways to connect.
The product extends beyond an embedded website widget, with integrations for Messenger and Instagram. Each channel brings its own delivery and conversation constraints, while the team still needs a coherent place to work.
The current chat runtime runs its agent logic inside the application. PostgreSQL stores application records, Redis supports coordination, and Qdrant supports knowledge retrieval. This is application engineering around AI: interfaces, state, background work, and integrations all have to agree.
AI is one part of the service.
Hello Marky Chat reflects how I like to approach an AI product. The model matters, but so do the handoff, the source material, and what happens after the answer. A polished chat bubble cannot make up for a broken workflow.
The result is a conversation product with room for both automation and human judgment. The assistant helps with the repeatable parts; the team can stay involved where its knowledge and care make the difference.
Good answers are useful. A connected conversation is better.
WHAT’S UNDERNEATH
Tools behind the experience.
- Next.js
- Node.js
- Express
- PostgreSQL
- Redis
- Qdrant