DineAssist — Every table. Every greeting.

YOUR DINING ROOM, IN A CONVERSATION

A little clarity. Just ask.

Available tables, service questions, and transparent wait estimates.

Table-aware assistant

✦

DineAssistYour floor companion

DineAssist

Hi! I can help you understand this dining room. Ask which tables are available, check a specific table, or explore an estimated wait.

Answers describe the current demo. Estimates are assumptions, not promises.

General questions and recent conversation context are sent to Twinly with table data. Don’t include personal information. The assistant cannot make bookings or change tables.

Floor map

14 mapped tables

Select a table to see its status and try a service action.

Your demo activity

Resets on refresh

Select a table and try recording a greeting.

COMPUTER VISION IN ACTION

From camera footage to table observations.

This is the saved detector output from our AI-generated sample restaurant video. Boxes show detected people; outlines show manually mapped table zones. The interactive floor demo uses separate illustrative service statuses.

Staff roles require manual identification. Person detection alone cannot confirm a greeting or a cleaned table.

LESS CHECKING. MORE CARING.

A second pair of eyes for the restaurant team.

A table sits uncleared. A guest waits for a greeting. A family waits at the door. DineAssist helps the team see where attention is needed next.

Map the room

Draw table zones on a camera image using OpenCV.

Track people

YOLO26 (Ultralytics, released January 2026) detects people. Tracking connects observations across frames.

Flag delays

A state machine tracks occupancy, greeting timers, and cleaning needs.

Help the team

The dashboard brings table status and service actions together.

Built with

Python · OpenCV · YOLO26 · FastAPI · SQLite · Streamlit

The full local application supports video and RTSP input, manual staff labels, recorded events, a waitlist, and host-triggered phone notifications. This Vercel site is a standalone HTML/CSS/JavaScript evaluation demo, with no server or camera connection.

Why YOLO26

YOLO26 was released on January 14, 2026, within the past year. It is designed for edge devices and drops the separate NMS post-processing step. On our sample video it ran about 15% faster than YOLO11 on a CPU and matched people to 10 of 14 tables as reliably; it found people at the smallest, farthest tables less often, so we lowered the confidence threshold to 0.25.

What comes next

Real restaurant testing, more reliable staff identification, and validated reporting on greeting delays and table turnaround. Predicting departure times and peak-hour reports are future work.