Skip to main content

​
AI Coach

Estudio Three’s AI Coach is your personal academic mentor—an intelligent assistant that understands your schedule, workload, and goals to provide contextual advice, study strategies, and motivational support.

​
What It Does

The AI Coach uses a large language model (Groq’s Llama3) to have conversations with you about:
  • Daily and weekly planning
  • Study strategies for specific subjects
  • Motivation and mindset coaching
  • Time management optimization
  • Recovery and break recommendations

Contextual Awareness

Coach knows your tasks, grades, habits, schedule, and Pomodoro history

Natural Conversations

Chat naturally about any academic or productivity challenge

Quick Actions

Pre-built prompts for common needs (plan day, get motivated, etc.)

Personalized Advice

Recommendations tailored to your sport, subjects, and energy levels

​
Why It Matters

Student-athletes face unique challenges that generic productivity advice doesn’t address:
  • Balancing competing demands (training vs studying)
  • Energy management across physical and mental tasks
  • Variable schedules with competitions and travel
  • Isolation from typical academic support systems
  • Time constraints that make traditional tutoring difficult
The AI Coach solves this by:
  • Being available 24/7 whenever you need guidance
  • Understanding the student-athlete context specifically
  • Providing instant feedback without judgment
  • Learning from your actual patterns and preferences
  • Offering strategies proven effective for your situation

​
How to Use It

​
Starting Your First Conversation

  1. Navigate to AI Coach page
  2. Choose a quick action button, or
  3. Type your question in the input field
  4. Click Send or press Enter
  5. Coach responds with personalized advice
  6. Continue the conversation as needed
The AI Coach has full context on your profile, active tasks, today’s Pomodoros, habits, grades, and upcoming events. The more data you have in the app, the better the advice.

​
Quick Action Prompts

Pre-built prompts for common needs:

Plan Day

Get a structured plan for today based on your schedule and tasks

Plan Week

Receive strategic weekly planning advice with priority allocation

Motivate Me

Get motivational boost tailored to your current situation

Break Advice

Learn optimal break strategies based on your training load

​
Example Conversations

You: Plan my dayCoach: Hey! Let’s structure your day. I see you have Physics at 9am, training at 3pm, and 3 active tasks.Here’s my recommendation:
  • 7:00-8:30am: Morning routine + breakfast
  • 8:30-9:00am: Review Physics notes before class
  • 9:00-12:00pm: Classes
  • 12:00-1:00pm: Lunch + rest
  • 1:00-2:45pm: Focus session - Math homework (2 Pomodoros)
  • 3:00-5:00pm: Training
  • 5:00-6:00pm: Recovery + dinner
  • 6:00-7:30pm: Focus session - English essay (3 Pomodoros)
  • 7:30-9:00pm: Review tomorrow’s materials
  • 9:00-10:00pm: Wind down routine
Your cognitive load is moderate today. You can handle 5 Pomodoros after training. Focus on Math first while energy is highest.
You: I’m struggling with Physics. What should I do?Coach: I see Physics is rated as difficulty 10 and your current average is 72%. Let’s build a recovery plan:
  1. Identify weak areas: Review your recent test. Which concepts did you miss?
  2. Active recall: Don’t just re-read. Practice problems without notes first.
  3. Spaced repetition: Study Physics for 1 Pomodoro daily, not cramming.
  4. Office hours: Schedule time with your teacher this week.
  5. Study group: Find 2-3 classmates to work through problems together.
I’ll prioritize Physics in your smart routine. Let’s aim for 5 hours this week. You’ve got this!
You: I’m exhausted and don’t want to studyCoach: I hear you. You completed training today and already did 2 Pomodoros this morning. That’s solid work!Remember why you’re doing this: [references your goals if set]. You’re not just a student or just an athlete—you’re proving you can do both.Here’s what I suggest:
  1. Take a real 20-minute break. No phone, just rest.
  2. Do ONE Pomodoro on your easiest task.
  3. Reassess. If still exhausted, it’s okay to stop.
Consistency beats perfection. One focused session today keeps your streak alive. Tomorrow is a new opportunity.

​
Types of Questions to Ask

Planning Questions:
  • “How should I structure my week?”
  • “What should I prioritize today?”
  • “Help me plan for exam week”
  • “I have a competition Saturday, how do I adjust?”
Study Strategy Questions:
  • “What’s the best way to study for [subject]?”
  • “How can I improve my grade in [subject]?”
  • “I have 3 exams next week, how do I prepare?”
  • “What study method works for [topic]?”
Motivation Questions:
  • “I’m feeling overwhelmed”
  • “How do I stay consistent?”
  • “I missed my goals this week”
  • “How do other student-athletes manage this?”
Tactical Questions:
  • “Should I study before or after training?”
  • “How long should my breaks be?”
  • “When is the best time to study [subject]?”
  • “How do I recover faster between tasks?”

​
Technical Details

​
AI Architecture

The AI Coach uses a serverless architecture: Client → Vercel API Proxy → Groq API → Llama3 Model

​
API Endpoint

Location: api/ai.ts The Vercel serverless function handles:
  • CORS configuration for web app
  • Rate limiting (30 requests per minute per IP)
  • Request validation and sanitization
  • Groq API communication
  • Error handling and fallbacks
Rate Limiting Implementation: 
const rateLimit = new Map<string, number[]>();

const ip = req.headers['x-real-ip'] || req.socket.remoteAddress;
const now = Date.now();
const times = rateLimit.get(ip) || [];
const validTimes = times.filter(t => now - t < 60000); // last minute

if (validTimes.length >= 30) {
  return res.status(429).json({ 
    error: "Too many requests", 
    code: 429 
  });
}

validTimes.push(now);
rateLimit.set(ip, validTimes);

​
Context Building

Before sending to the LLM, the client builds rich context: Location: src/features/ai-coach/components/AICoachPage.tsx:31 
function buildContext() {
  return {
    userProfile: {
      name: user.name,
      sport: user.mainSport,
      waterGoal: user.waterGoal
    },
    dailyRoutine: {
      totalCognitiveLoad: routine.cognitiveLoad,
      totalPhysicalLoad: routine.physicalLoad
    },
    language: settings.language,
    activeTasks: tasks.filter(t => t.status !== 'completed'),
    todayPomodoros: pomodoroStore.sessionsCompleted,
    habits: habits.map(h => ({
      name: h.title,
      streak: h.currentStreak
    })),
    subjects: subjects.map(s => ({
      name: s.name,
      average: academicsStore.getWeightedAverage(s.id),
      nextExam: s.nextExamDate
    })),
    upcomingEvents: calendar.getUpcomingEvents(7),
    totalResources: resources.length
  };
}
This context is sent with every message, allowing the AI to provide personalized responses.

​
System Prompt

The AI Coach uses a carefully crafted system prompt: Location: api/ai.ts:79 
const systemPrompt = `Eres Routine Optimizer AI Coach. Ayudas a estudiantes-deportistas.
Contexto del usuario:
- Nombre: ${context.userName}
- Deporte principal: ${context.mainSport}
- Tareas activas: ${context.activeTasks.join(', ')}
- Pomodoros de hoy: ${context.todayPomodoros}
Responde en ${context.language}, sé directo, motivador y claro.`;
This tells the LLM:
  • Its role (student-athlete coach)
  • User’s specific situation
  • Current state (tasks, progress)
  • Communication style (direct, motivating)

​
Model Selection

Using Groq’s Llama3-8B model:
  • Fast inference (< 1 second response time)
  • Strong reasoning capabilities
  • Cost-effective for high volume
  • Supports multi-language responses
API Call (api/ai.ts:87): 
const groqResponse = await fetch('https://api.groq.com/openai/v1/chat/completions', {
  method: 'POST',
  headers: {
    'Authorization': `Bearer ${process.env.GROQ_API_KEY}`,
    'Content-Type': 'application/json'
  },
  body: JSON.stringify({
    model: 'llama3-8b-8192',
    messages: [
      { role: 'system', content: systemPrompt },
      { role: 'user', content: userMessage }
    ]
  })
});

​
State Management

Store: src/stores/useChatStore.ts 
interface ChatMessage {
  id: string;
  role: 'user' | 'assistant';
  content: string;
  timestamp?: string;
}

interface ChatState {
  messages: ChatMessage[];
  isLoading: boolean;
  addMessage: (message: ChatMessage) => void;
  clearMessages: () => void;
  setLoading: (loading: boolean) => void;
}
Messages are stored in local state only (not persisted to database). Each session starts fresh.

​
Components

AICoachPage (src/features/ai-coach/components/AICoachPage.tsx:16)
  • Main chat interface
  • Message history display
  • Input field with send button
  • Quick action buttons
  • Loading states
Message Display:
  • User messages: Right-aligned, primary color
  • Assistant messages: Left-aligned, muted background
  • Avatar icons distinguish speakers
  • Rounded corners and smooth animations

​
Error Handling

Multiple layers of error handling:
  1. Client-side validation (src/features/ai-coach/services/aiService.ts:27)
    • Checks for empty messages
    • Validates context structure
    • Handles network errors
  2. API-level validation (api/ai.ts:49)
    • Validates request body
    • Checks message length (max 2000 chars)
    • Enforces rate limits
  3. Graceful degradation
    • 429 (Too many requests): “You’ve sent too many messages…”
    • 503 (Service unavailable): “AI service unavailable…”
    • Network error: “Connection error. Check your internet.”

​
Security & Privacy

Rate Limiting

30 requests per minute per IP prevents abuse and controls costs

Input Sanitization

All user inputs sanitized before sending to AI to prevent injection

No Message Storage

Chat history not saved to database—privacy by default

CORS Protection

API endpoint restricted to app domain only

​
API Key Security

Groq API key stored as environment variable:
  • Never exposed to client
  • Accessed only in serverless function
  • Rotated periodically
  • Monitored for usage patterns

​
Best Practices

Better: “I have a Physics exam Friday and I’m weak on thermodynamics. How should I study?”Worse: “Help me study”The more context you provide, the better the advice.
If you’re asking a common question (plan day, get motivated), use the quick action buttons. They’re optimized prompts that trigger specialized responses.
The AI Coach is only as good as your data. Keep tasks, grades, and habits current for best advice.
Have conversations, not just one-off questions. Follow up with “Why?” or “Can you elaborate?” to dig deeper.

​
Limitations

Current Limitations:
  • Chat history not persisted (resets on page refresh)
  • No long-term memory across sessions
  • Cannot directly modify your schedule or create tasks
  • Limited to text-based advice (no visual diagrams)
  • Rate limited to prevent abuse (30 requests/minute)
Future Enhancements:
  • Persistent conversation history
  • Proactive suggestions based on patterns
  • Direct integration (“Schedule this for me”)
  • Voice input/output support
  • Multi-modal responses (charts, schedules)

​
Integration with Other Features

  • Task Management: Coach sees active tasks and can suggest priorities
  • Pomodoro Timer: Uses today’s focus time in recommendations
  • Academic Tracking: Knows your grades and can suggest study focus
  • Habits & Goals: References your goals for motivational context
  • Smart Routines: Understands your schedule and energy patterns
  • Calendar: Aware of upcoming events and deadlines

​
Language Support

The AI Coach responds in your configured language:
  • English
  • Spanish (default)
  • Portuguese
  • French
Set language in Settings → Language. The coach automatically adapts.
Next Steps: Explore the complete Core Features overview to see how all features work together, or dive into Getting Started to begin your journey.