Boss Assistant

Why it was built

Human operators repeatedly spend attention on status follow-ups, progress reports, and routine information requests. Static chatbots cannot answer reliably when operational data changes continuously.

The goal is not to remove human communication, but to reserve human attention for situations where judgment and direct communication add the most value.

How Boss Assistant works

1. The assistant classifies the request and determines whether it can answer safely.
2. Supported requests retrieve current operational data from Google Sheets or grounded knowledge from the RAG layer.
3. The assistant generates a response based on that retrieved information rather than a static script.
4. Ambiguous, sensitive, unsupported, or high-touch cases are escalated to a real person.

Supported work

• Answer questions about live work and project status.
• Handle routine follow-up, reporting, and progress requests.
• Collect requirements and help estimate requested changes.
• Review supported attachments as part of the conversation.
• Escalate requests that require human judgment or direct communication.

Human-in-the-loop design

The assistant is designed around grounded retrieval rather than unrestricted generation. It handles routine requests first, but uncertainty is a reason to escalate?not a reason to invent an answer.

My role

As system designer and reviewer, I designed the workflow, connected the assistant to dynamic operational data and RAG knowledge, defined the escalation boundary, reviewed the AI-assisted implementation, and validated the private internal deployment.