“To know thyself is to predict thyself.” TiresiasIQ is an experimental AI engine that learns your daily behavior patterns and predicts future actions from natural language input. Inspired by the mythic oracle Tiresias, this project transforms your everyday logs into actionable foresight using neural networks.
Yes you heard it right, TiresiasIQ is a Human Behavior Prediction Engine, the first of it's kind to be ever created.
It's on GitHub and will very soon be made available to your phones and computers.
There are fragments, attempts, and adjacent projects—but no open-source system or AI engine fully tailored to personal behavior prediction based on daily habit logging and time-window-based neural estimation with natural language input and action forecasting as you’re building in TiresiasIQ.
Feature | TiresiasIQ | Existing Systems |
---|---|---|
Daily self-logging of tasks | ✅ | ❌ (mostly passive tracking) |
Action completion window prediction (e.g. 2 hours) | ✅ | ❌ |
Uses neural nets (FFN, LSTM) for personal action forecasting | ✅ | ❌ |
Natural Language Interpretation of Tasks | ✅ (v2) | ❌ (very rare or too general) |
Tailored to one individual for personal feedback loop | ✅ | ❌ (most are generalized) |
CLI logger + full dashboard with predictions | ✅ | ❌ |
Related Projects (but fundamentally different) include
- Google Timeline + Activity Recognition What it does: Logs where you go, what you do on phone.
Why it's different: It recognizes, not predicts, and it’s closed-source.
- Replika / AI Companions What they do: Chat-based behavior adaptation.
Why it's different: They respond emotionally, not forecast rational behavior based on past logs.
- Apple / Fitbit HealthKit + Wellness AI What they do: Predict when to stand, walk, sleep, etc.
Why it's different: Predicts biological rhythm, not cognitive decision/action-based tasks.
- Habitica / Streak Apps What they do: Habit gamification, track whether you did something.
Why it's different: No real AI prediction. Just behavior encouragement.
- Smart Personal Assistants (e.g., Siri Shortcuts, Alexa Routines) What they do: Suggest actions at certain times/locations.
Why it's different: Hardcoded patterns. No neural learning, no contextual understanding.
- nudge.ai, x.ai (now defunct) What they did: Predict best time to contact people, send reminders.
Why it's different: Built for business & CRM, not personal task-life modeling.
- Academic Works Like: “Forecasting Personal Behaviors from Mobile Data” (MIT) Uses phone sensor data (e.g., calls, location) to predict future behavior.
But it's passive data, very coarse granularity.
No daily self-logging or natural language.
- "MyBehavior" (Cornell Tech) Recommends health activities based on logs.
Doesn’t handle custom tasks or task completion prediction.
TiresiasIQ is a first-of-its-kind open system that combines:
- Daily user-logged data,
- Real-time and time-window forecasting,
- Neural network-based prediction,
- Natural language task processing
Architecture Overview
Frontend: Streamlit GUI for entering tasks, viewing predictions.
Core Model:
- FFN-LSTM hybrid
- Plans for Conv-LSTM enhancement (semantic + temporal co-pattern learning)
Input features:
- Extracted keywords and action verbs from natural language
- Sentiment polarity and subjectivity
- Timestamps normalized by hour and weekday
- spaCy vector embeddings (action semantics)
Stack
- TensorFlow, Keras
- spaCy, TextBlob
- pandas, scikit-learn
- Streamlit
- SQLite (behavior.db)
Example Prediction
"Will I finish writing the essay tonight?" → Confidence: 76% | Interpreted action: "write"
Roadmap
- Build FFN-LSTM baseline
- Add NLP-powered semantic extraction
- Implement Conv1D-LSTM hybrid
- WebSocket API for external clients
- Exportable personal behavior fingerprint
- Self-discovery insights (e.g., procrastination patterns, success-hour maps)