.jpg)
Process description
This automation is an intelligent power management system via the Telegram interface. The system automatically identifies new users, conducts onboarding to calculate personal BJU take, keeps a food diary, analyzes food photos counting calories and macronutrients, monitors progress and provides personalized dietary recommendations.
API keys and services:
- Telegram Bot API - for the user interface (bot: Nutrition)
- OpenAI API Key - for AI agents and image analysis (gpt-4.1, gpt-4.1-mini, gpt-4o)
- Google Gemini API - backup language model
- Google Sheets OAuth2 - for storing user data and a food diary
- SerPapi - to find information about products and dishes
System architecture by blocks
SECTION 1: RECEIVING AND ROUTING REQUESTS
1.1 Telegram Trigger — Entry Point
Purpose: Tracks all incoming messages and photos from users Settings:
- Updates: message
- Credentials: Nutrition
1.2 New User Verification
Purpose: Checks if a user is in the Google Sheets database Settings:
- Document ID: (Goals table)
- Sheet: Sheet1
- Return First Match: true
- Credentials: Your credentials
1.3 Aggregate - Processing results
Purpose: Aggregates verification data to determine user status Settings:
- Field to Aggregate: ID
- Aggregation Type: standard
1.4 If - Conditional Routing
Purpose: Directs users to the appropriate processing flow Condition:
- Condition: $json.id contains $ ('Telegram Trigger') .item.json.message.chat.id
- True Output: Existing User → Switch
- False Output: New User → AI Agent1
1.5 Switch - Content Classification
Purpose: Splits incoming messages into text and photos Conditions:
- Output 0 “Text”: message.text exists
- Output 1 “Photo”: message.photo exists
SECTION 2: ONBOARDING NEW USERS
2.1 AI Agent1 - Nutrition Program Creation Agent
Purpose: Onboards new users and creates a personalized nutrition program
System prompt:
# Nutrition Program Creation Agent - Onboarding
## Your role
You're a specialist in creating personalized nutrition programs. You work ONLY with new users.
## Work scenario
### Message 1: Welcome
“Hi! 👋 Pleased to meet you!
I'll help you create your personalized nutrition program in 3 minutes.
What is your name?”
### Message 2: After the name
“Nice to meet you [Name]!
Tell us about yourself:
- Gender and age?
- Height and weight?”
### Message 3: After receiving data
“Great! Now let's define the goal:
1 ️ ⃣ Lose weight
2 ️ ⃣ Gain weight
3 ️ ⃣ Support the form
What do you choose? (write a number)”
### Message 4: After choosing a goal
[If losing weight]: “How many kilos do you want to lose weight?”
[If set]: “How many kg do you want to gain?”
[If maintained]: move on to the next
### Message 5: Activity Level
“The last question is your activity:
1 ️ ⃣ Sedentary lifestyle
2 ️ ⃣ 1-2 workouts per week
3 ️ ⃣ 3-4 workouts per week
4 ️ ⃣ 5+ workouts per week
Choose a number”
### Message 6: Calculate and Save
[Calculate using formulas]
[Call SAVE_USER_PROGRAM]
“✨ Done, [Name]!
📊 Your program:
- [XXXX] kcal per day
- Proteins: [X] g
- Fat: [X] g
- Carbohydrates: [X] g
The program has been saved! Now you can text me what you eat and I'll count everything 💪”
## Formulas
BMR male = (10×weight) + (6.25×height) - (5×age) +5
BMR women = (10×weight) + (6.25×height) - (5×age) -161
TDEE = BMR × [1.2/1.375/1.55/1.725]
Calories = TDEE ± goal adjustment
Connected components:
- OpenAI Chat Model1 (gpt-4.1-mini)
- Simple Memory1 (Session Key: chat.id)
- SAVE_USER_PROGRAM - program saving tool
2.2 SAVE_USER_PROGRAM - Save a nutrition program
Purpose: Saves the new user's personal nutrition program to Google Sheets Settings:
- Operation: AppendorUpdate
- Matching Column: ID
- Fields to be filled in by an AI agent:
- ID (from Telegram)
- First name (Name)
- Gender
- Age
- Weight (Weight)
- Height
- Desired result (Result)
- Physical. Activity
- Goal
- Norma BJU (Daily goals)
SECTION 3: WORKING WITH EXISTING USERS
3.1 AI Agent — Principal Nutrition Agent
Purpose: Works with existing users, keeps food records and makes recommendations
System prompt (short version):
# Principal Nutrition Agent
## Your role
You're working with users who ALREADY HAVE a nutrition program. You help keep track of food and monitor progress.
## Main scenario
### With any food report:
1. **Get context right away: **
- GET_USER_PROGRAM (user plan)
- GET_USER_MEAL (what's already eaten today)
2. **Analyze food: **
- If you need data about KBJU → GET_INFO
- Calculate calories and BJU
3. **Reply to user: **
“[Dish name] is a great choice!
📊 Nutritional value:
- [XXX] kcal
- Proteins: [X] g
- Fat: [X] g
- Carbohydrates: [X] g
📈 Your day:
Eats: [XXX]/[normal] kcal ([X]%)
The goal is left: [XXX] kcal
[Advice based on progress]
Am I recording it?”
4. **Upon confirmation: **
- SAVE_USER_MEAL
Connected components:
- OpenAI Chat Model (gpt-4.1) - basic model
- Google Gemini Chat Model - backup model
- Simple Memory (Your chat ID)
- GET_USER_PROGRAM - getting a nutrition program
- GET_USER_MEAL - getting a food diary
- SAVE_USER_MEAL - preserving food intake
- GET_INFO - search for product information
3.2 AI Transform — Data preparation
Purpose: Unifies the data format from different sources (text/photo) for an AI agent Settings:
- Converts all input data to the “Text” parameter
- Keeps the contents of message.text if it exists
- JavaScript code for transformation
SECTION 4: FOOD PHOTO PROCESSING
4.1 Get a file - Upload photos from Telegram
Purpose: Downloads a food photo sent by a user Settings:
- Resource: file
- File ID: $ ('Telegram Trigger') .item.json.message.photo [2] .file_id
- Credentials: Nutrition
4.2 Analyze image - AI image analysis
Purpose: Analyzes food photos and determines the composition and calorie content
Instructions for analysis:
MAIN FEATURES
- When viewing an image of a food, identify each food and its main components (proteins, carbohydrates, fats, etc.).
- Use standard dimensions (e.g. 26 cm plate, 250 ml cup) to determine the scale.
- Pay attention to whether the dish looks like the one prepared in the restaurant — if so, add extra fat to cook.
- Estimate portion sizes in grams. Use the guidelines in the image.
- Make realistic assumptions. Give preference to normal portion sizes.
- Evaluate calories and macronutrients for each product using reliable databases.
- Calculate calories for each food by entering the likely range.
- Summarize your total calorie range.
Settings:
- Model: gpt-4o
- Input Type: base64
- Text: + caption instructions from the user
SECTION 5: DATA TOOLS
5.1 GET_USER_PROGRAM - Getting a nutrition program
Purpose: Extracts the user's personal nutrition program from Google Sheets Settings:
- Document: “Goals” table
- Filter: ID = chat.id
- Fields: all fields of the nutrition program
5.2 GET_USER_MEAL - Getting a food diary
Purpose: Retrieves the user's meal records for the current day Settings:
- Document ID: 1v2wcfRwakesBMIBAMC4YZ3ZFMY90ZGSHGFEWXDQuiq0
- Filter: ID = chat.id
- Fields: date, time, dish, KBJU, portion weight
5.3 SAVE_USER_MEAL - Saving meals
Purpose: Records a new meal in a food diary Settings:
- Operation: append
- AI fields to fill in:
- Date and time (automatic)
- User ID
- Dish
- Calories
- Proteins
- Carbs (carbs)
- Fats
- Portion weight (weight of food)
- Admission category (type of serving)
5.4 GET_INFO - Information search
Purpose: Searches information about products and dishes via SerpAPI Settings:
- Tool Type: SerPapi
- Query: is formed by an AI agent
- Credentials: Your credentials
SECTION 6: SENDING RESPONSES
6.1 Send a text message - Reply to existing users
Purpose: Sends responses from the main AI agent to users Settings:
- Chat ID: $ ('Telegram Trigger') .item.json.message.chat.id
- Text: $json.output
- AppendAttribution: false
6.2 Send a text message1 - Reply to new users
Purpose: Sends responses from an onboarding agent to new users Settings: Identical to Send a text message
Node connection diagram
Main stream:
Telegram Trigger → New User Verification → Aggregate → If
New User Thread:
If (False: new) → AI Agent1 → Send a text message1
Existing user thread:
If (True: Existing) → Switch
├─ Switch (Text) → AI Transform → AI Agent → Send a text message
└─ Switch (Photo) → Get a file → Analyze image → AI Transform → AI Agent → Send a text message
Connecting AI components to the main agent:
OpenAI Chat Model ──
Gemini Chat Model ───→ AI Agent
Simple Memory ──────
GET_USER_PROGRAM ───
GET_USER_MEAL ──────
SAVE_USER_MEAL ────
GET_INFO ─────────
Onboarding agent connections:
OpenAI Chat Model1 ─
Simple Memory1 ────→AI Agent1
SAVE_USER_PROGRAM ──
Required services and their settings
Telegram Bot setup:
- Create a bot via @BotFather
- Get the token and save it as “Nutrition” credentials
- Set up rights to receive messages and photos
- Install commands: /stats, /today, /left
Setting up Google Sheets:
- Create two sheets in Google Sheets:
- “Goals” table for nutrition programs
- “Nutrition” table for a food diary
- Set up OAuth2 for Google Sheets
- The structure of the “Goals” table:
- ID, Name, Gender, Age, Weight, Height, Desired Result, Physical Activity, Goal, BJU Norm
- The structure of the “Nutrition” table:
- ID, Date, Time, Dish, Calories, Protein, Carbs, Fat, Portion Weight, Intake Category
Setting up AI services:
- OpenAI API:
- Models: gpt-4.1, gpt-4.1-mini, gpt-4o
- Set up billing for all models
- Google Gemini API:
- Backup model for the main agent
- SerPapi:
- To find product information
System capabilities
Features for new users:
- Interactive onboarding in 3 minutes
- Calculation of the KBJU personal rate using scientific formulas
- Taking into account individual goals (sliming/set/maintenance)
- Adapting to activity levels
Features for existing users:
- Keeping a food diary via text or photo
- AI analysis of food photos with the definition of the composition
- Counting calories and macronutrients in real time
- Tracking daily progress
- Personalized recommendations based on goals
- Searching for information about all foods and dishes
AI capabilities:
- Multimodality: working with text and images
- Contextual memory: personalization for every user
- Intelligent analysis: identifying products and services from photos
- Adaptive recommendations: advance-based tips
Use cases
Scenario 1: Onboarding a new user
🤖 Bot: Hi! 👋 Pleased to meet you! What is your name?
👤 User: Anna
🤖 Bot: Nice to meet you, Anna! Tell us about yourself...
[dialogue continues]
🤖 Bot: ✨ Done! Your norm: 1850 kcal
Scenario 2: Accounting for food through text
👤 User: I ate a burger with potatoes
🤖 Bot: Potato burger is a great choice!
📊 Nutritional value: 820 kcal
📈 Eats: 1320/1850 kcal (71%)
Am I writing it down?
👤 User: Yes
🤖 Bot: ✅ Recorded!
Scenario 3: Food photo analysis
👤 User: [sends a photo of the salad]
🤖 AI analysis: [determines composition]
🤖 Bot: Caesar salad is a great choice!
📊 Approximately 380 kcal
The goal is left: 150 kcal
Scenario 4: Statistics and Recommendations
👤 User: /stats
🤖 Bot: 📊 Weekly statistics:
- Average consumption: 1790 kcal
- Goal completed: 5 days out of 7
- Weight trend: -0.8 kg
Great progress! Keep it up!
System application
For personal use:
- Nutrition control without complex applications
- Fast food accounting via photo or text
- Personal program taking into account the goals
- Motivation by tracking progress
For fitness trainers:
- Customer monitoring through a common database
- Automating a routine CBJU calculations
- Objective control meals for wards
- Scaling practitioners
For nutritionists:
- Remote support patients
- Detailed analytics on nutrition
- Increased commitment to the program
- Save time at consultations
The result of the system
What happens is:
- Personal AI nutritionist in your pocket
- Automated food diary with AI analysis
- Goal Achievement System by weight
- Knowledgebase about products and dishes
- Motivational support 24/7
Performance metrics:
- Onboarding time: 3 minutes
- Photo analysis accuracy: 85-90%
- Processing speed: 5-10 seconds
- User commitment: 70% active after a month
- Achieving goals: 65% achieve their goal in 3 months
System advantages:
- Simplicity: messenger interface instead of complex applications
- Personalization: an individual program for everyone
- Flexibility: working with text and photos
- Accessibility: works on any Telegram device
- Science: calculations using proven formulas
ROI and practical value:
- Savings on a nutritionist
- Save time: 5 minutes instead of 30 to keep a diary
- Improving performance: +40% goal achievement
- Scalability: unlimited users