HerDay AI — A Voice Companion for Emotional Reflection

A voice-based AI companion that helps women reflect on their day, track emotional patterns, and receive gentle weekly well-being insights

Comment

Inspiration

Many women silently carry emotional weight every day — balancing work, family, expectations, and personal responsibilities. While physical health is openly discussed, emotional well-being often goes unnoticed.

We realized that many women don’t always have a safe, uninterrupted space to talk about their feelings. Journaling takes effort and consistency. Therapy can feel intimidating or inaccessible. Sometimes, all someone needs is a listener.

We asked ourselves: What if emotional reflection could feel as natural as having a conversation?

That idea led to HerDay AI — a voice-based emotional companion designed to listen without judgment and gently support daily reflection.

What it does

HerDay AI is a voice-based AI companion that allows users to:

Speak freely about their day

Automatically detect emotional tone using NLP

Receive empathetic, emotionally aware responses

Log reflections with timestamps

Track emotional patterns over time

View emotion trends by day (Monday–Sunday)

Generate intelligent behavioral feedback

Instead of typing journal entries, users simply press a button and talk. The system:

Converts speech to text

Detects emotional state (joy, sadness, anger, fear, etc.)

Generates a warm, human-like AI response

Stores the emotion and reflection for pattern analysis

Users can also request deeper feedback to understand emotional habits such as complaining patterns, blame tendencies, or signs of growth.

HerDay AI is not a therapist or diagnostic tool — it is a reflective companion that encourages emotional awareness.

How we built it

We built HerDay AI as an end-to-end AI pipeline combining speech processing, emotion detection, and large language models.

Speech-to-Text: We used Whisper to convert voice input into text.

Emotion Detection: We integrated a transformer-based model (j-hartmann/emotion-english-distilroberta-base) to classify emotional tone.

Mathematically, emotion classification estimates:

P(emotion∣reflection) P(emotion∣reflection)

AI Response Generation: We used Groq’s high-speed inference API with the llama-3.1-8b-instant model to generate empathetic, emotionally intelligent replies.

Emotion Logging System: Each reflection is stored with date, day, time, detected emotion, and text. This enables long-term pattern tracking.

Analysis Engine: We built endpoints that:

Calculate dominant emotion

Compute positivity vs negativity ratios

Generate structured psychological feedback using LLM reasoning

Frontend was built using HTML, CSS, and JavaScript with:

MediaRecorder API (voice input)

SpeechSynthesis API (voice output)

Fetch API for backend communication

Backend was built using FastAPI.

Challenges we ran into

Making AI feel emotionally natural Early responses felt robotic. We refined prompts carefully to maintain warmth and honesty.

Avoiding medical boundaries We ensured the system never gives clinical advice or diagnoses.

Emotion detection limitations Human emotions are complex. The classifier sometimes simplifies overlapping feelings.

Deployment constraints Running heavy speech models on backend made deployment challenging. We explored architecture improvements for scalability.

Accomplishments that we're proud of

Built a complete voice-to-LLM pipeline

Integrated real-time emotion detection

Designed an intelligent emotional feedback engine

Implemented day-wise emotion tracking

Created structured behavioral analysis

Designed voice-interactive UI with control options

Most importantly, we created a system that feels emotionally intentional — not just technically impressive.

What we learned

Through this project, we learned:

AI empathy depends heavily on prompt engineering.

Emotional pattern tracking is more powerful than single-day analysis.

Responsible AI design requires ethical boundaries.

Emotional awareness tools must prioritize psychological safety.

We also learned that technology should amplify reflection — not replace human connection.

What's next for HerDay AI — A Voice Companion for Emotional Reflection

In the future, we plan to:

Add visual dashboards with emotion charts

Implement weekly automated emotional summaries

Integrate vector memory for long-term reflection recall

Add multilingual support (including Hindi)

Replace CSV logging with scalable databases

Deploy globally using optimized architecture

Strengthen privacy and authentication mechanisms

Our long-term vision is to make HerDay AI a trusted emotional reflection companion that supports women in building healthier emotional habits over time.

Built With

  • csv
  • distilroberta-(emotion-detection)
  • fastapi
  • groq
  • html/css
  • hugging-face-transformers
  • javascript
  • llama-3.1-8b
  • mediarecorder-api
  • python
  • web-speech-api
  • whisper-(speech-to-text)
Share this project:

Updates