title	emoji	colorFrom	colorTo	sdk	sdk_version	app_file	pinned
RideCastAI2.0	🚕	red	yellow	streamlit	1.35.0	ui/app.py	true

🚕 RideCastAI 2.0 — Real-Time Fare & ETA Predictor with Drift Recovery and Latency Optimization

🚦 RideCastAI 2.0 is a production-grade real-time ML system that predicts ride fares & ETAs, detects input/output drift, updates models live, and tracks latency — simulating a dynamic dispatch engine like Uber, Lyft, or Grab.

🎯 Designed for top-tier AI infrastructure roles where adaptability, performance, and monitoring are key.

---

⚡️ What’s New in 2.0

Feature	1.0	🚀 2.0 Upgrade
Prediction Scope	Static fare & ETA prediction	🔁 Real-time prediction with async ingestion
Drift Detection	None	✅ Input + Output drift with visual alerts
Model Adaptation	Offline model	✅ Online learning with river
Caching & Inference	None	✅ ONNX + Joblib caching + asyncio
Latency Tracking	No	✅ Real-time latency chart
Error Debugging	No	✅ Live top-5 worst predictions
UI	Basic chart	✅ Tabbed Streamlit dashboard with toggles
Deployability	Basic	✅ Free-tier deployable on Hugging Face

---

📊 Architecture Overview

Rides (CSV/Simulated) 
→ Feature Preprocessing
→ Model Inference (ONNX / River)
→ Caching Layer (Joblib)
→ Output Metrics Tracker (MAE / RMSE / Drift)

↳ Input Drift Detector (HalfSpaceTrees)
↳ Output Drift Detectors (KSWIN, ADWIN)
↳ Online Learner (River LinearRegression)
↳ Latency Tracker (Async + Retry logic)

↳ Streamlit UI with Live Dashboard:
   - Prediction + Drift + Latency Tabs
   - Drift Injection / Online Update Toggle
   - Top-5 Error Viewer
   - Logs + Trace Overlay

---

🌟 Key Features

✅ Real-Time Fare & ETA Prediction
✅ Online Learning (River) — self-adapts to incoming data
✅ Input Drift Detection (HalfSpaceTrees)
✅ Output Drift Detection (KSWIN, ADWIN)
✅ ONNX-Optimized Model + Joblib Caching
✅ Live Top-5 Worst Prediction Viewer
✅ Latency Visualization per Batch
✅ Drift Injection Toggle for Testing
✅ Fully Modular Streamlit UI
✅ Free-tier Deployable (No LLMs required)

---

🧠 Component Stack

Each module is modular, production-grade, and latency-aware.

Component	Role	Tech Stack
🚕 Predictor	Predicts fare & ETA	scikit-learn, onnx, river
🧠 Drift Detectors	Input (HalfSpaceTrees), Output (KSWIN/ADWIN)	river.drift
♻️ Online Learner	Updates model weights per ride	river.linear_model.LinearRegression
💾 Caching	Stores past predictions for reuse	joblib
🕒 Latency Tracker	Logs inference time and averages	time, asyncio
📈 Visual Overlay	RMSE, Drift Flags, Top-5 Errors	matplotlib, seaborn
🧪 Drift Injector	Force anomaly to test system recovery	Manual + Config toggle
🖥️ UI Layer	Live dashboard with tabs	Streamlit

---

🔎 How It Works

• Ingests a real-time ride (or simulated stream)
• Featurizes and routes through predictor
• Monitors:
  • 🚨 Input Drift via HalfSpaceTrees
  • 📉 Output Drift via KSWIN & ADWIN
• Updates model weights online (if enabled)
• Caches previous rides to reduce inference cost
• Logs latency, RMSE, drift spikes, and error outliers
• Visualizes all data on an elegant Streamlit UI

---

🖼️ UI Preview

Tabbed layout includes:

• 🔮 Live Prediction — Real-time results + top-5 worst errors
• ⚠️ Drift & Metrics — Input + Output drift tracking + RMSE overlay
• ⚡ Latency Monitor — Inference timing graph
• 🔧 Settings — Inject Drift, Enable Online Learning, View Architecture

📌 Try the live app here:
👉 RideCastAI 2.0 – Hugging Face Space

---

🧪 Sample Output (Ride ID: ride_1027)

• 🚕 Predicted Fare: ₹184.76
• 🚕 Predicted ETA: 12.4 minutes
• 📊 RMSE (Last 50): Fare: ₹9.12 | ETA: 1.8 mins
• 📉 Drift:
  • Input: ❌ No
  • Output: ✅ ADWIN Triggered
• Latency: 1.2s
• In Top-5 Error: ✅ Yes → Underestimated ETA by 4.9 mins

---

📁 Project Structure

RideCastAI2.0/
├── ui/
│   └── app.py                     # Streamlit frontend
├── model/
│   └── river_models.py            # Drift + model logic
├── utils/
│   ├── latency_tracker.py
│   └── drift_plot_utils.py
├── data/
│   └── rides.csv                  # Input data
├── assets/
│   ├── ridecast_architecture.png  # Arch Diagram
│   └── ridecast_demo.gif          # UI Demo
└── requirements.txt

---

💼 Why This Project Stands Out

• ✅ Real-time architecture, not just static ML
• ✅ Combines drift detection + online learning + latency awareness
• ✅ Debuggable like internal Uber/Lyft tools
• ✅ Clean, modular UI with tabbed monitoring
• ✅ Designed for free-tier deployability with zero cost
• ✅ Built as a candidate portfolio app to demonstrate elite ML engineering

---

🧰 Run Locally (for Devs)

git clone https://github.com/rajesh1804/RideCastAI2.0
cd RideCastAI2.0
pip install -r requirements.txt
streamlit run ui/app.py

---

🧠 Linked Projects

Project	Description	Link
🧵 ThreadNavigatorAI 2.0	Multi-Agent Reddit thread analyzer with LLM-as-a-Judge	🔗 View
🛒 GroceryGPT+	Vector search + reranking grocery assistant with fuzzy recall	🔗 View
🎬 StreamWiseAI	Netflix-style movie recommender with Retention Coach Agent	🔗 View

---

⚠️ Known Challenges

• 🧱 river failed to install on Hugging Face (Rust required) → solved via downgrade to river==0.15.1
• 🌀 Drift injection needed careful UI/UX isolation — solved with toggle + sidebar logs
• 🧠 ONNX conversion limited to scikit-learn baseline only (not River) — handled fallback with hybrid logic
• 🕒 Async scheduler in Streamlit was tricky — solved using asyncio.create_task + stateful cache

---

🧑‍💼 About Me

Rajesh Marudhachalam — AI/ML Engineer building real-time, agentic, and production-grade AI systems.
📍 GitHub | LinkedIn

Projects: ThreadNavigatorAI2.0, StreamWiseAI, GroceryGPT+

---

🙌 Acknowledgments

• River — Streaming ML + Drift Detection
• Streamlit — UI Framework
• Hugging Face Spaces — App Hosting

---

📜 License

MIT License

⭐️ Star this repo if it impressed you. Follow for more production-grade ML builds.