Go Eats

A food recommender system using a content-based filter ML model. This project consists of many components as part of a microservice architecture:

Component	Repo
Telegram Bot	/botservice
ML Model	/recommendationservice
Form + Admin Dashboard	go-eats-form repo
REST Web Server	go-eats-server repo
Refer to the README in each folder/repo for more developer details

NOTE: The system is offline from 12 midnight to 11am due to budget constraints

To demo: telegram bot or search for @goeats_bot in Telgram!

Table of Contents

• About
  • Motivation
  • Features
  • Tech Stack
  • Architecture
• Screenshots
• Setup
• Plan
  • Scope
  • User stories
  • Phases
  • Planned Features
  • Deployment

About

Motivation

This project came about the problem of deciding for a food place as a group. Due to undecisiveness, only vague preferences for a food place is decided like cuisine dont-wants, budget, etc and not the actual place. Go Eats solves this problem by recommending a food place based on these vague preferences.

Features

1. Telegram bot for easy access which queries for preferences
2. Machine learning based recommendations
3. Form for user submission of reviews for data collection
4. Admin portal for approval and preprocessing of reviews

Tech Stack

Component	Tech
Telgram Bot	Golang, gRPC
ML Model	Python, scikit-learn, gRPC
Form + Dashboard	Next.js, React, TailwindCSS
Web Server	Golang, REST API
Deployment	Docker, Kubernetes, AKS, Terraform
Database	Azure Database for PostgreSQL

Architecture

Microservices Architecture with Kubernetes was adopted

Screenshots

TBA

Setup

Clone this repo Refer to individual README for specific setups

Set .env variables:

DOCKER_PASSWORD=<dockerhub password for image repo>

For dockerizing, some commands:

docker-compose build <name(optional)>
docker-compose push <name(optional)>
docker-compose up <name(optional)>

docker-compose deployment names:

• recommendation-service
• bot-service

Project Plan

Scope

Telegram bot - Get user preferences for food, recommends places

• based on mealtime (breakfast, brunch, lunch or dinner)
• don't-wants, cuisines the user would not want
• budget
• mood (comfort, energy, indulgent, healthy, adventurous(out-of-user prefs))

OpenAI API - get dish recommendations

PostgreSQL - To store restaurants and reviews

• Places:
  • name
  • budget
  • moods
  • cuisine types
  • location
  • summary
• Reviews:
  • name
  • review

Recommendation Algorithm

• Data collection
  • Manual entry (import from current spreadsheet)
  • Web scraper and Google Maps API
  • Allow users to recommend restaurants through a form, pending approval and edits from admin
• Model
  • Content-based filter
  • Vectorize text TF-IDF on reviews to recommend similar restaurants
• Server
• Improvements
  • periodically retrain model by fetching data from DB

Web form - allow users to submit restaurant reviews

• Name of place (auto complete with Google Maps API) only in SG
• isChain (nullifies location parameter under the hood)
• Average budget
• Mood dropdown single select
• Cuisine dropdown multiple
• Mealtime dropdown multiple
• Rating stars out of 5

Admin Dashboard

• pending table
  • able to edit, delete and approve
• approved table
  • able to edit and delete
  • pagination
  • sort by column

User stories

• As an indecisive user I want to get food recommendation based on my mood and budget
• As an adventurous person I want to get food outside my comfort zone
• As a meticulous person I want to see which dish is recommended so that I can maximise my budget

Phase

Version 1.0 [DONE]

• Telegram bot <-> Server
• OpenAI API
• Suggest food places through prompt engineering

Version 2.0 [DONE]

• Telegram bot - Server - DB
• Form/Dashboard - Server - DB
• Store restaurants
• Suggest food based on DB
• Allow users to add restaurants to a website (mobile based) and wait for pending approval
• Admin accepts/rejects/edits restaurant requests to the DB

Version 3.0 [DONE]

• Telegram bot(gRPC client) - Server(ML, gRPC server) - DB
• ML model Recommendation Algorithm (content-based filtering)
• Package the model as a microservice
• gRPC

Version 3.1

• QOL updates
  • Telegram bot webhooks
  • refine inputs for telegram bot prompts (allow for anything)
  • TLS for web server
  • Admin portal security
• Methods for updating model
  • query from DB

Planned Features

1. telegram input menu

• mealtime (brunch, lunch, dinner, snack)
• cuisine dont wants
• budget ()
• mood (comfort, healthy, indulgent, casual, fancy, anything)

2. user recommendations

• submit personal recommendations through a website
• pending approval from admin

3. generate output based on similar food place

Deployment

DB: Azure Database for PostgreSQL - Flexible

Host Kubernetes cluster on AKS

• Size should be limited (use minimal plan)
• Automate shutting down and running the cluster using Azure automation

Managed with Terraform IaC