Answer Ratings

Improve future answer suggestions by rating generated answers.

Overview

Answer ratings help Intervu learn your preferences. Good examples are used to guide future generations, while poor examples are avoided.

---

Rating Answers

How to Rate

Each answer card has two buttons:

• Thumbs Up 👍 — This answer was helpful
• Thumbs Down 👎 — This answer was not helpful

Click either button to register your rating.

What Happens When You Rate

When you rate an answer:

1. The rating is stored locally
2. Rated answers are added to your LLM context
3. Future prompts include "good examples" and "bad examples"

---

How Ratings Improve Answers

In-Context Learning

When you rate answers, the LLM receives context:

Good Examples:

Here are examples of answers the user found helpful:
Q: What is your experience with React?
A: I've been working with React for5 years, starting with class components and transitioning to hooks. Most recently, I led a team building a real-time dashboard that handles 100K+ concurrent users...

Bad Examples:

The user found these answers unhelpful — avoid this style:
Q: Tell me about yourself.
A: I am a software engineer. I like coding. I have many skills.

Improved Responses

With ratings, the LLM:

• Matches the style of your good examples
• Avoids patterns from bad examples
• Becomes more aligned with your preferences over time

---

Rating Guidelines

When to Rate

You don't need to rate every answer. Rate strategically:

✅ Rate thumbs up when:

• Answer structure is clear and logical
• Content directly answers the question
• Examples from your resume are used appropriately
• Tone matches interview context

✅ Rate thumbs down when:

• Answer is too vague or generic
• Content doesn't match your experience
• Structure is confusing
• Answer is factually incorrect

❌ Don't rate when:

• Answer is mediocre (neutral)
• You edited the answer significantly
• Answer is still generating

Consistency Matters

Rate consistently to help the system learn:

• Same style = same rating
• Don't randomly rate
• Focus on what helps you

---

Managing Ratings

View Rated Answers

Rated answers are stored locally but not displayed in the UI.

Clear All Ratings

To start fresh:

1. Open Settings (gear icon)
2. Scroll to Advanced Settings
3. Click Clear Ratings
4. Confirm the action

This removes all stored ratings.

Clear Speaker Data

When speaker diarization is enabled, you can also clear speaker data:

1. Click the Clear menu (trash icon) in the title bar
2. Select Clear speaker data
3. This resets speaker names and embeddings while keeping transcript

Or use Clear transcript & answers which also clears speaker data.

Export Ratings

Ratings are stored in:

Windows:

%APPDATA%/intervu/ratings.json

macOS:

~/Library/Application Support/intervu/ratings.json

You can back up this file to preserve your learning data.

---

Technical Details

Storage

Ratings are saved as:

[
  {
    "id": "qa-1234567890-abcde",
    "question": "What is your experience with React?",
    "answer": "I've been working with React for...",
    "rating": "up",
    "timestamp": 1234567890000
  }
]

Context Window

The last 3 good examples and last 2 bad examples are included in each LLM request.

This keeps:

• Context focused on recent preferences
• Token usage reasonable
• Response time fast

---

Best Practices

Over Time

Session	Good Examples	Bad Examples
1	0	0
5	3	2
10	5	3
20+	3-5 (rotating)	2

Ratings accumulate and rotate — older ratings are replaced by newer ones.

Interview Types

Different interview types may need different rating patterns:

Technical Interview:

• Rate answers that demonstrate depth
• Focus on code examples and architecture

Behavioral Interview:

• Rate answers with good STAR structure
• Focus on storytelling narrative

Leadership Interview:

• Rate answers that show strategic thinking
• Focus on business impact

When to Clear Ratings

Clear ratings when:

• Changing job types (frontend → backend)
• Changing experience level (junior → senior)
• Starting a new resume completely

---

FAQ

Do ratings sync across devices?

No, ratings are stored locally. Each device has its own rating history.

Will ratings make answers identical?

No, ratings guide style — each answer is still generated fresh based on the current question.

Can I edit ratings after the fact?

Currently, no. You can only clear all ratings and start fresh.

Does rating affect performance?

No, ratings are loaded once per session. There's no performance impact.

---

Next Steps

• Basic Mode — Standard workflow
• Speaker Diarization — Multi-speaker identification
• Interview Timer — Track interview duration
• Contextual Tips — Key points extraction