Janis
A self-healing AI system that manages my fitness, nutrition, and academics — all through one conversation.
The Problem
I started with three separate AI bots — one for fitness, one for nutrition, one for school. Each worked fine on its own. But they couldn't talk to each other. A hard workout didn't adjust my calorie targets. Exam stress didn't change my training plan. I was the one carrying context between three separate conversations.
I needed one system. One conversation. One AI that could think across every part of my life.
The Architecture
I designed the system as a house. Not just a metaphor — the actual architecture maps to rooms, doors, and utilities.
What It Does
Academics
Connected to my university's Canvas system, the study service knows my class schedule, upcoming assignments, and exam dates. After each lecture, it turns my notes into a quick quiz to reinforce what I just learned.
When homework is assigned, it reads the problems from the textbook PDF, generates a blank worksheet for my iPad, and creates a separate tutor guide with step-by-step solutions. Both get uploaded to Google Drive automatically.
- Canvas API integration for live assignment tracking
- Post-lecture quizzes generated from my notes
- Homework packets with LaTeX math rendering
- Tutor guides with worked solutions
- Google Drive upload with organized folders
Nutrition
I can snap a photo of my meal and send it to Janis. She identifies the food using AI vision, estimates the macros, and logs everything automatically. Over time, she learns my regular meals — "the usual burrito" becomes a shortcut.
Throughout the day, she tracks my running totals against targets: 2,300 calories, 168g protein (non-negotiable), and coaches me on what I need before the day ends.
- Photo-based food logging with AI vision
- Real-time macro tracking with progress bars
- Favorites system that learns my regular meals
- Daily nutrition logs with meal-by-meal breakdown
- Proactive coaching on protein and calorie pacing
Fitness
The fitness service knows my training schedule, tracks progressive overload, and logs every workout with exercises, sets, reps, and weights. It's connected to Apple Health for step counts, sleep scores, and active energy.
It celebrates PRs, flags when I'm under-sleeping, and adjusts recommendations based on what the rest of the system knows — like whether I have a big exam tomorrow or if I'm behind on protein.
- Apple Health integration for steps, sleep, and activity
- Progressive overload tracking with PR detection
- Mile time progression tracking (11:00 → 7:30 goal)
- Weekly training reviews with volume analysis
- Cross-domain awareness — adapts to stress and nutrition
Self-Healing
Every night at 3:30 AM, Janis runs 16 tests on herself — checking if each service responds, if the data is valid, if the AI pipeline works, if external APIs are connected.
If something fails, she doesn't just report it. She kills the broken service, restarts it, and tests again. If it passes, she marks it as "healed" and moves on. If not, she messages me on Telegram with a diagnosis.
I wake up to either silence (everything's fine) or a specific problem and what she already tried. I never wake up to something that's been quietly broken for hours.
Morning Briefing
Every day at 5:15 AM, Janis sends a briefing to my phone. The weather. The top 3 news stories curated from 7 RSS feeds. My schedule for the day. Any system issues from overnight.
By the time I'm brushing my teeth, I already know what today looks like.
Built With
Want the Full Story?
I wrote about the entire journey — from three broken bots to a self-healing AI house — on Substack.