Lentera

Inspiration

We kept noticing the same pattern in the people around us — friends, classmates, ourselves. Nobody realizes they're burning out until they've already hit a wall. Mood slips gradually, so a downward trend is nearly impossible to spot in yourself, and by the time someone finally reaches out for help, they've usually been struggling for months. Professional care is expensive, waitlists are long, and stigma keeps people quiet.

We wanted to build the emotional equivalent of a step counter: a small, private daily habit that quietly notices when things are heading in the wrong direction and gently points you toward support early — before it becomes a crisis. We called it Lentera, which means "lantern": a small light to help you find your way in the dark.

What it does

Lentera turns a 60-second daily journal check-in into an early-warning system for your mental health. Each day you write a few words about how you're feeling and rate your mood. Behind the scenes, a machine-learning model reads the emotional tone of your entry, converts it into a wellbeing score, and tracks it over time on a simple trend dashboard.

When it detects a sustained downward trend, Lentera responds appropriately:

• Coping micro-actions on a hard day (breathing, grounding, reaching out to someone).
• An early-warning nudge when your mood has been declining for several days.
• Immediate crisis resources when your writing signals acute risk — pointing you toward a real person, because Lentera never tries to replace professional care.

It's private by default and takes under a minute, so it's easy enough to actually stick with.

How we built it

We built the core app in Python with Streamlit for a fast, interactive web interface. The emotional intelligence comes from a Hugging Face Transformers model (j-hartmann/emotion-english-distilroberta-base), which classifies each journal entry across emotions like sadness, fear, anger, and joy. We convert that distribution into a single wellbeing score, blend it with the user's self-rated mood, and store it as a time series.

A lightweight trend-detection algorithm (built with pandas) compares recent check-ins against the user's baseline to flag sustained declines, and a response engine decides whether to show coping tips, an early-warning message, or crisis resources based on severity. We also built a static HTML/CSS/JavaScript prototype with Chart.js for a fast, zero-install demo, and used Git/GitHub for version control with GitHub Pages hosting the prototype.

Challenges we ran into

The hardest part wasn't the code — it was the tone. In mental health, wording matters enormously, so we spent real effort making every message feel supportive, never clinical or alarming. Getting the trend detection right meant balancing sensitivity against false alarms: too twitchy and it cries wolf, too dull and it misses the moment that matters. And handling crisis language responsibly was a serious design challenge — we had to make sure the model's limitations never put a user at risk, which is why we added a keyword safety net and always route to human help rather than trying to handle a crisis ourselves.

Accomplishments that we're proud of

We're proud that Lentera actually works — a real ML model reading genuine journal text and producing a meaningful, visible mood trend, not just a mockup. We're proud of the responsible design: a mental-health tool that knows its own limits and points people to real help instead of overreaching. And we're proud that we kept it genuinely simple — a full check-in takes 60 seconds, which is the difference between an app people abandon and one they actually use.

What we learned

We learned how much design and empathy matter in a technical project — the same feature can help or harm depending entirely on how it's worded. We got hands-on experience turning a pre-trained Transformer model into a real product feature: taking raw emotion scores and shaping them into something smoothed, trustworthy, and actionable. And we learned that responsible AI in a sensitive domain means building in humility — designing the system to escalate to humans rather than pretending to solve everything itself.

What's next for Lentera

• Habit-building tools — optional check-in reminders and streaks so the daily habit sticks.
• Shareable summaries — opt-in reports users can bring to a counselor or doctor, turning journaling into real clinical signal.
• Regional crisis directories — verified, location-aware hotlines and resources.
• Deeper personalization — learning each user's baseline over time for more accurate, individualized early warnings.
• More languages — so Lentera can reach the communities that most wellness apps ignore.

Built With

• css
• html
• javascript
• python
• transformers