Upload folder using huggingface_hub

README.md

@@ -12,8 +12,8 @@ pinned: false
   <p><strong>Cardiovascular Telemetry & Edge MLOps Engine</strong></p>
   
   <p>
-    <a href="https://huggingface.co/spaces/IDKHowToCodeFR/HEARTFLOW_OS" target="_blank">
-      <img src="https://img.shields.io/badge/🤗_HuggingFace-Live_Demo-FFD21E.svg?style=for-the-badge" alt="Hugging Face Space" />
+    <a href="https://idkhowtocodefr-tinyml-backend.hf.space/docs" target="_blank">
+      <img src="https://img.shields.io/badge/🤗_HuggingFace-API_Docs-FFD21E.svg?style=for-the-badge" alt="Hugging Face API" />
     </a>
     <a href="https://idkhowtocodefr.github.io/HEARTFLOW_OS" target="_blank">
       <img src="https://img.shields.io/badge/🌐_Website-Live_App-black.svg?style=for-the-badge&logo=githubpages" alt="Live Website" />

backend/database.py

@@ -91,8 +91,9 @@ async def init_db() -> None:
 
 async def log_prediction(data: Any, prediction_label: str, confidence: float) -> None:
     async with aiosqlite.connect(DB_PATH) as db:
-        from datetime import timezone, timedelta
-        ist = timezone(timedelta(hours=5, minutes=30))
+        from datetime import datetime
+        import zoneinfo
+        ist = zoneinfo.ZoneInfo("Asia/Kolkata")
         await db.execute('''
             INSERT INTO predictions (timestamp, heart_rate, spo2, sys_bp, dia_bp, temp, fall_detection, prediction_label, confidence)
             VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)

backend/main.py

@@ -2,17 +2,14 @@ import sys
 import os
 sys.path.append(os.path.dirname(os.path.abspath(__file__)))
 
-from fastapi import FastAPI, UploadFile, File, BackgroundTasks, HTTPException, WebSocket, WebSocketDisconnect, Depends
+from fastapi import FastAPI, UploadFile, File, BackgroundTasks, HTTPException, WebSocket, WebSocketDisconnect
 from fastapi.middleware.cors import CORSMiddleware
-from pydantic import BaseModel
 import pandas as pd
 import asyncio
-import random
 import shap
 import database as db
-import numpy as np
+import logging
 from contextlib import asynccontextmanager
-from config import settings
 import warnings
 
 # Suppress sklearn version mismatch warnings in stdout
@@ -21,11 +18,20 @@ try:
     warnings.filterwarnings("ignore", category=InconsistentVersionWarning)
 except ImportError:
     pass
+
 from preprocessing import preprocess_data
 from export import generate_c_code
 from inference import evaluate
+from schemas import PatientData
+from mlops import MLOpsEngine
+from streamer import TelemetryStreamer
+from fastapi.responses import RedirectResponse
+
+# Setup standard logging for critical fault alerts
+logging.basicConfig(filename="alerts.log", level=logging.WARNING, 
+                    format='%(asctime)s %(levelname)s: %(message)s', datefmt='%Y-%m-%d %H:%M:%S')
 
-ensemble_system = None
+ml_engine = None
 
 async def background_sync():
     while True:
@@ -35,20 +41,16 @@ async def background_sync():
 
 @asynccontextmanager
 async def lifespan(app: FastAPI):
-    global ensemble_system
+    global ml_engine
     await db.init_db()
-    
     sync_task = asyncio.create_task(background_sync())
     
-    try:
-        from ensemble import EnsembleModel
-        ensemble_system = EnsembleModel()
-        print("Models loaded successfully.")
-    except Exception as e:
-        print(f"Model loading failed: {e}")
+    data_path = os.path.join(os.path.dirname(os.path.abspath(__file__)), '..', 'data', 'patient_dataset.csv')
+    ml_engine = MLOpsEngine(data_path)
+    
     yield
     sync_task.cancel()
-    ensemble_system = None
+    ml_engine = None
 
 app = FastAPI(title="TinyML Healthcare API", lifespan=lifespan)
 
@@ -60,36 +62,13 @@ app.add_middleware(
     allow_headers=["*"],
 )
 
-def get_ensemble():
-    global ensemble_system
-    return ensemble_system
-
-class PatientData(BaseModel):
-    Heart_Rate: float
-    SpO2_Level: float
-    Systolic_BP: float
-    Diastolic_BP: float
-    Body_Temp: float
-    
-from fastapi.responses import RedirectResponse
-
 @app.get("/")
 def read_root():
     return RedirectResponse(url="/docs")
 
 @app.get("/health")
 def health_check():
-    return {"status": "Healthy" if get_ensemble() else "Warning - Models Offline"}
-
-def log_alert_sync(data: PatientData, conf: float):
-    try:
-        from datetime import datetime, timezone, timedelta
-        ist = timezone(timedelta(hours=5, minutes=30))
-        timestamp = datetime.now(ist).strftime("%Y-%m-%d %H:%M:%S")
-        with open("alerts.log", "a") as f:
-            f.write(f"[{timestamp}] ALERT: Patient at risk! HR: {data.Heart_Rate}, SpO2: {data.SpO2_Level}, Confidence: {conf:.2f}\n")
-    except Exception:
-        pass
+    return {"status": "Healthy" if ml_engine and ml_engine.get_ensemble() else "Warning - Models Offline"}
 
 async def log_prediction_background(data, final_pred, conf):
     try:
@@ -100,13 +79,11 @@ async def log_prediction_background(data, final_pred, conf):
 @app.post("/predict")
 async def predict(data: PatientData, background_tasks: BackgroundTasks):
     try:
-        eng = get_ensemble()
+        eng = ml_engine.get_ensemble()
         if not eng:
             return {"error": "Models untrained. Ensure python backend/models.py executes."}
                 
-        from inference import evaluate
         result = await asyncio.to_thread(evaluate, eng, data)
-        
         if "error" in result:
             return result
             
@@ -114,11 +91,11 @@ async def predict(data: PatientData, background_tasks: BackgroundTasks):
         final_pred = result["prediction_label"]
         conf = result["probability"]
         
-        # Critical Alert System — fault-tolerant
+        # Critical Alert System — standard library logging
         if is_at_risk == 1 and float(conf) > 0.80:
-            background_tasks.add_task(log_alert_sync, data, float(conf))
+            logging.warning(f"Patient at risk! HR: {data.Heart_Rate}, SpO2: {data.SpO2_Level}, Confidence: {conf:.2f}")
                 
-        # Log to SQLite History — fault-tolerant
+        # Log to SQLite History
         background_tasks.add_task(log_prediction_background, data, final_pred, float(conf))
         
         return result
@@ -129,34 +106,15 @@ async def predict(data: PatientData, background_tasks: BackgroundTasks):
 @app.websocket("/ws/feed")
 async def websocket_feed(websocket: WebSocket):
     await websocket.accept()
-    
-    eng = get_ensemble()
+    eng = ml_engine.get_ensemble()
     if not eng:
         await websocket.close(code=1011)
         return
         
-    from simulator import PatientDataSimulator
-    simulator = PatientDataSimulator()
-    
+    streamer = TelemetryStreamer(eng)
     try:
-        async for state in simulator.run():
-            data = PatientData(**state)
-            
-            prediction_result = await asyncio.to_thread(evaluate, eng, data)
-            
-            front_pred = None
-            if "error" not in prediction_result:
-                front_pred = {
-                    "is_at_risk": 1 if prediction_result["prediction"] else 0,
-                    "label": prediction_result["prediction_label"],
-                    "confidence": prediction_result["probability"],
-                    "disease_probs": prediction_result.get("disease_probs", {})
-                }
-            
-            await websocket.send_json({
-                "sensor_data": data.model_dump(),
-                "prediction": front_pred
-            })
+        async for payload in streamer.stream():
+            await websocket.send_json(payload)
     except WebSocketDisconnect:
         pass
     except Exception as e:
@@ -168,11 +126,9 @@ async def history():
 
 @app.get("/dataset")
 def get_dataset():
-    data_path = os.path.join(os.path.dirname(os.path.abspath(__file__)), '..', 'data', 'patient_dataset.csv')
-    if os.path.exists(data_path):
+    if ml_engine and os.path.exists(ml_engine.dataset_path):
         try:
-            # Force UTF-8 and strip column whitespace
-            df = pd.read_csv(data_path, encoding='utf-8')
+            df = pd.read_csv(ml_engine.dataset_path, encoding='utf-8')
             df.columns = [c.strip() for c in df.columns]
             return df.to_dict(orient="records")
         except Exception as e:
@@ -187,18 +143,11 @@ def force_sync():
 
 @app.post("/explain")
 async def explain(data: PatientData):
-    eng = get_ensemble()
+    eng = ml_engine.get_ensemble()
     if not eng or 'rf' not in eng.models:
         return {"error": "RF Model unavailable for explanation."}
         
-    df = pd.DataFrame([{
-        'Heart Rate (bpm)': data.Heart_Rate,
-        'SpO2 Level (%)': data.SpO2_Level,
-        'Systolic Blood Pressure (mmHg)': data.Systolic_BP,
-        'Diastolic Blood Pressure (mmHg)': data.Diastolic_BP,
-        'Body Temperature (°C)': data.Body_Temp
-    }])
-    
+    df = pd.DataFrame([data.model_dump()])
     X_proc, _ = await asyncio.to_thread(preprocess_data, df, False)
     
     def compute_shap():
@@ -223,56 +172,22 @@ async def explain(data: PatientData):
 
 @app.get("/export_tinyml")
 def export_tinyml(model_name: str = "rf", quantize: bool = False):
-    eng = get_ensemble()
+    eng = ml_engine.get_ensemble()
     return generate_c_code(eng, model_name, quantize)
 
-def run_training_background():
-    try:
-        from models import train_models
-        train_models()
-        # Force reload in lifespan is tricky, but we can do it via global
-        global ensemble_system
-        from ensemble import EnsembleModel
-        ensemble_system = EnsembleModel()
-    except Exception as e:
-        print(f"Background training failed: {e}")
+def _run_mlops_ingest(csv_bytes: bytes):
+    return ml_engine.ingest_batch_sync(csv_bytes)
 
 @app.post("/retrain")
 async def retrain(background_tasks: BackgroundTasks, file: UploadFile = File(...)):
-    data_path = os.path.join(os.path.dirname(os.path.abspath(__file__)), '..', 'data', 'patient_dataset.csv')
-    os.makedirs(os.path.dirname(data_path), exist_ok=True)
+    content = await file.read()
     
-    try:
-        content = await file.read()
-        import io
-        new_df = pd.read_csv(io.BytesIO(content))
-        new_df.columns = [c.strip() for c in new_df.columns]
-        
-        if os.path.exists(data_path):
-            existing_df = pd.read_csv(data_path, encoding='utf-8')
-            existing_df.columns = [c.strip() for c in existing_df.columns]
-            
-            # Validate schema
-            required_cols = set(existing_df.columns)
-            provided_cols = set(new_df.columns)
-            
-            if not required_cols.issubset(provided_cols):
-                missing = required_cols - provided_cols
-                return {"error": f"Schema mismatch. Missing columns: {list(missing)}"}
+    # We add this to background tasks so the HTTP response returns immediately
+    # while the engine merges CSVs and retrains the models.
+    def bg_task():
+        res = ml_engine.ingest_batch_sync(content)
+        if not res.success:
+            print(f"MLOps Background Task Failed: {res.message}")
             
-            # Ensure columns are in the same order and select only necessary ones
-            new_df = new_df[existing_df.columns]
-            
-            # Append data
-            combined_df = pd.concat([existing_df, new_df], ignore_index=True)
-        else:
-            combined_df = new_df
-            
-        # Save validated and combined dataset
-        combined_df.to_csv(data_path, index=False, encoding='utf-8')
-        
-        background_tasks.add_task(run_training_background)
-        
-        return {"status": "success", "message": f"Dataset updated (now {len(combined_df)} records) and ensemble retrain started in background!"}
-    except Exception as e:
-        return {"error": f"Retraining failed: {str(e)}"}
+    background_tasks.add_task(bg_task)
+    return {"status": "success", "message": "Dataset uploaded and ensemble retrain started in background!"}

backend/mlops.py

@@ -0,0 +1,64 @@
+import os
+import io
+import pandas as pd
+from typing import Optional
+
+class RetrainResult:
+    def __init__(self, success: bool, message: str):
+        self.success = success
+        self.message = message
+
+class MLOpsEngine:
+    """
+    Deep Module encapsulating the MLOps Lifecycle:
+    - Maintains the active ensemble model.
+    - Handles schema validation and data ingestion natively with pandas.
+    - Orchestrates model retraining and hot-swapping.
+    """
+    def __init__(self, dataset_path: str):
+        self.dataset_path = dataset_path
+        self.active_ensemble = None
+        self._load_active_models()
+
+    def _load_active_models(self):
+        try:
+            from ensemble import EnsembleModel
+            self.active_ensemble = EnsembleModel()
+        except Exception as e:
+            print(f"MLOps: Model loading failed (likely untrained): {e}")
+
+    def get_ensemble(self):
+        return self.active_ensemble
+
+    def ingest_batch_sync(self, csv_bytes: bytes) -> RetrainResult:
+        try:
+            new_df = pd.read_csv(io.BytesIO(csv_bytes))
+            new_df.columns = [c.strip() for c in new_df.columns]
+            
+            if os.path.exists(self.dataset_path):
+                existing_df = pd.read_csv(self.dataset_path, encoding='utf-8')
+                existing_df.columns = [c.strip() for c in existing_df.columns]
+                
+                # Check for schema mismatch before concatenation
+                missing = set(existing_df.columns) - set(new_df.columns)
+                if missing:
+                    return RetrainResult(False, f"Schema mismatch. Missing columns: {list(missing)}")
+                
+                # Ponytail shrink: pandas join="inner" drops mismatched extra columns cleanly
+                combined_df = pd.concat([existing_df, new_df], join="inner", ignore_index=True)
+            else:
+                combined_df = new_df
+                
+            os.makedirs(os.path.dirname(self.dataset_path), exist_ok=True)
+            combined_df.to_csv(self.dataset_path, index=False, encoding='utf-8')
+            
+            # Re-train models synchronously
+            from models import train_models
+            train_models()
+            
+            # Hot-swap the live models
+            self._load_active_models()
+            
+            return RetrainResult(True, f"Dataset updated (now {len(combined_df)} records) and ensemble hot-swapped!")
+        except Exception as e:
+            return RetrainResult(False, f"Retraining failed: {str(e)}")

backend/schemas.py

@@ -0,0 +1,8 @@
+from pydantic import BaseModel
+
+class PatientData(BaseModel):
+    Heart_Rate: float
+    SpO2_Level: float
+    Systolic_BP: float
+    Diastolic_BP: float
+    Body_Temp: float

backend/streamer.py

@@ -0,0 +1,36 @@
+import asyncio
+from typing import AsyncGenerator
+from simulator import PatientDataSimulator
+from inference import evaluate
+from schemas import PatientData
+
+class TelemetryStreamer:
+    """
+    Deep Module for Telemetry Streaming.
+    Encapsulates the simulator lifecycle, ML inference threading, and payload formatting.
+    Provides a simple async generator interface.
+    """
+    def __init__(self, ensemble_model):
+        self.ensemble_model = ensemble_model
+        self.simulator = PatientDataSimulator()
+
+    async def stream(self) -> AsyncGenerator[dict, None]:
+        async for state in self.simulator.run():
+            data = PatientData(**state)
+            
+            # Offload ML inference to thread pool to avoid blocking the event loop
+            prediction_result = await asyncio.to_thread(evaluate, self.ensemble_model, data)
+            
+            front_pred = None
+            if "error" not in prediction_result:
+                front_pred = {
+                    "is_at_risk": 1 if prediction_result.get("prediction") else 0,
+                    "label": prediction_result.get("prediction_label"),
+                    "confidence": prediction_result.get("probability"),
+                    "disease_probs": prediction_result.get("disease_probs", {})
+                }
+            
+            yield {
+                "sensor_data": data.model_dump(),
+                "prediction": front_pred
+            }

frontend/src/app/architecture/page.tsx

@@ -4,152 +4,172 @@ import { motion } from "framer-motion";
 
 const easeFluid: [number, number, number, number] = [0.16, 1, 0.3, 1];
 
-const staggerContainer = {
-  hidden: { opacity: 0 },
-  show: {
-    opacity: 1,
-    transition: { staggerChildren: 0.1, delayChildren: 0.1 }
-  }
-};
-
 const fadeUp = {
-  hidden: { opacity: 0, y: 20 },
-  show: { opacity: 1, y: 0, transition: { duration: 0.8, ease: easeFluid } }
+  hidden: { opacity: 0, y: 10 },
+  show: { opacity: 1, y: 0, transition: { duration: 0.6, ease: easeFluid } }
 };
 
-const BlueprintNode = ({ title, desc, pulse = false }: { title: string, desc: string, pulse?: boolean }) => (
-  <motion.div 
-    whileHover={{ x: 5 }}
-    className="border border-slate/30 bg-ink/50 p-4 relative group hover:border-hazard transition-colors cursor-default"
-  >
-    {pulse && (
-      <div className="absolute right-4 top-1/2 -translate-y-1/2 flex items-center justify-center">
-        <motion.div 
-          animate={{ scale: [1, 2, 1], opacity: [0.8, 0, 0.8] }} 
-          transition={{ duration: 1.5, repeat: Infinity, ease: "easeInOut" }}
-          className="absolute w-3 h-3 bg-hazard rounded-full"
-        />
-        <div className="w-1.5 h-1.5 bg-hazard rounded-full relative z-10" />
-      </div>
-    )}
-    <div className="text-[10px] text-slate tracking-[0.1em] mb-1 uppercase">{desc}</div>
-    <div className="text-[14px] text-white font-bold tracking-tight uppercase">{title}</div>
-  </motion.div>
-);
-
 export default function Architecture() {
   return (
     <main className="min-h-[100dvh] bg-canvas-cream flex flex-col font-mono selection:bg-hazard selection:text-white pb-16 overflow-x-hidden">
-      <motion.div 
-        variants={staggerContainer}
-        initial="hidden"
-        animate="show"
-        className="max-w-[1200px] mx-auto w-full px-4 md:px-8 py-8 md:py-16 flex-1 flex flex-col"
-      >
-        <motion.header variants={fadeUp} className="mb-8 md:mb-12 border-b-4 border-ink pb-8 flex justify-between items-end">
+      
+      <div className="max-w-[1400px] mx-auto w-full px-4 md:px-8 py-16 flex-1 flex flex-col">
+        
+        {/* HEADER BLOCK */}
+        <motion.header 
+          initial="hidden" animate="show" variants={fadeUp}
+          className="mb-16 border-b-2 border-ink pb-8 flex flex-col md:flex-row justify-between items-start md:items-end gap-8"
+        >
           <div>
-            <div className="text-[12px] text-hazard font-bold tracking-[0.1em] mb-4">
-              /// DOCS_AND_SPECS
+            <div className="text-[10px] text-hazard font-bold tracking-[0.2em] mb-4 uppercase flex items-center gap-4">
+              <span className="w-2 h-2 bg-hazard animate-pulse" />
+              SYSTEM_ARCHITECTURE // DECLASSIFIED
             </div>
-            <h1 className="text-[clamp(2.5rem,6vw,6rem)] leading-tight">
-              SYSTEM<br />ARCHITECTURE
+            <h1 className="text-[clamp(3rem,8vw,8rem)] leading-[0.85] font-sans font-black tracking-[-0.04em] uppercase text-ink">
+              PIPELINE<br />TOPOLOGY
             </h1>
           </div>
+          <div className="text-[12px] text-slate font-bold tracking-[0.1em] uppercase text-right max-w-xs">
+            REF: HFL-OS-ARCH-01<br/>
+            REV: 2.6.4<br/>
+            STATUS: ACTIVE
+          </div>
         </motion.header>
 
-        {/* Live Hardware Blueprint */}
-        <motion.div variants={fadeUp} className="w-full bg-ink p-8 md:p-12 mb-16 relative font-mono overflow-hidden border-2 border-ink shadow-[8px_8px_0px_0px_rgba(242,56,39,1)]">
-          {/* Animated Scanning Line */}
-          <motion.div 
-            animate={{ top: ["-10%", "110%"] }}
-            transition={{ duration: 8, ease: "linear", repeat: Infinity }}
-            className="absolute left-0 right-0 h-32 bg-gradient-to-b from-transparent via-hazard/10 to-transparent pointer-events-none z-0"
-          />
-          
-          <div className="relative z-10 flex flex-col lg:flex-row gap-12 lg:gap-16 items-stretch">
+        {/* BLUEPRINT GRID */}
+        <motion.div 
+          initial="hidden" animate="show" variants={fadeUp}
+          className="w-full bg-ink grid grid-cols-1 md:grid-cols-3 gap-[2px] p-[2px] mb-16 shadow-[16px_16px_0px_0px_rgba(230,25,25,1)]"
+        >
+          {/* COLUMN 1 */}
+          <div className="bg-canvas-cream flex flex-col justify-between p-8 relative group">
+            <div className="absolute top-2 left-2 text-ink/30 text-[10px]">+</div>
+            <div className="absolute top-2 right-2 text-ink/30 text-[10px]">+</div>
+            <div className="absolute bottom-2 left-2 text-ink/30 text-[10px]">+</div>
+            <div className="absolute bottom-2 right-2 text-ink/30 text-[10px]">+</div>
             
-            {/* Stage 1: Ingestion */}
-            <div className="flex-1 flex flex-col gap-6 border-l-2 border-slate/30 pl-6 relative">
-              <div className="absolute -left-[5px] top-0 w-2 h-2 bg-hazard" />
-              <div className="text-[10px] text-hazard tracking-[0.2em] font-bold">PHASE_01 // INGESTION</div>
-              <BlueprintNode title="PATIENT SENSORS" desc="60Hz Telemetry Stream" pulse />
-              <BlueprintNode title="CLIENT UI" desc="Next.js 16 / React 19" />
-              <BlueprintNode title="CSV BATCH" desc="Offline Dataset Upload" />
+            <div className="text-[10px] text-hazard tracking-[0.2em] font-bold mb-8">PHASE_01 // INGESTION</div>
+            <div className="flex flex-col gap-12">
+              <div>
+                <h3 className="font-sans text-[24px] font-black uppercase text-ink tracking-tight mb-2 group-hover:text-hazard transition-colors">[ PATIENT SENSORS ]</h3>
+                <p className="text-[12px] text-slate uppercase leading-tight">60Hz Telemetry Stream via WebSockets</p>
+              </div>
+              <div>
+                <h3 className="font-sans text-[24px] font-black uppercase text-ink tracking-tight mb-2 group-hover:text-hazard transition-colors">[ CSV BATCH ]</h3>
+                <p className="text-[12px] text-slate uppercase leading-tight">Offline Dataset Upload for MLOps</p>
+              </div>
+              <div>
+                <h3 className="font-sans text-[24px] font-black uppercase text-ink tracking-tight mb-2 group-hover:text-hazard transition-colors">[ CLIENT UI ]</h3>
+                <p className="text-[12px] text-slate uppercase leading-tight">React 19 / Next.js 16 Edge Render</p>
+              </div>
             </div>
+            <div className="mt-16 text-[10px] text-slate uppercase border-t border-slate/30 pt-4">INPUT_VECTORS &gt;&gt;&gt;</div>
+          </div>
 
-            {/* Stage 2: Processing */}
-            <div className="flex-1 flex flex-col gap-6 border-l-2 border-slate/30 pl-6 relative">
-              <div className="absolute -left-[5px] top-0 w-2 h-2 bg-hazard" />
-              <div className="text-[10px] text-hazard tracking-[0.2em] font-bold">PHASE_02 // CORE_ROUTER</div>
-              <BlueprintNode title="FASTAPI SERVER" desc="Async WebSockets & REST" pulse />
-              <BlueprintNode title="SQLITE DB" desc="Aiosqlite Async Storage" />
+          {/* COLUMN 2 */}
+          <div className="bg-canvas-cream flex flex-col justify-between p-8 relative group">
+            <div className="text-[10px] text-hazard tracking-[0.2em] font-bold mb-8">PHASE_02 // CORE_ROUTER</div>
+            <div className="flex flex-col gap-12">
+              <div>
+                <h3 className="font-sans text-[24px] font-black uppercase text-ink tracking-tight mb-2 group-hover:text-hazard transition-colors">[ FASTAPI SERVER ]</h3>
+                <p className="text-[12px] text-slate uppercase leading-tight">Async Uvicorn Router & REST API</p>
+              </div>
+              <div className="border-l-4 border-hazard pl-4">
+                <h3 className="font-sans text-[24px] font-black uppercase text-ink tracking-tight mb-2 group-hover:text-hazard transition-colors">[ SQLITE DB ]</h3>
+                <p className="text-[12px] text-slate uppercase leading-tight">Aiosqlite Non-Blocking Storage</p>
+              </div>
             </div>
+            <div className="mt-16 text-[10px] text-slate uppercase border-t border-slate/30 pt-4">PROCESSING_BUS &gt;&gt;&gt;</div>
+          </div>
 
-            {/* Stage 3: Inference & Compilation */}
-            <div className="flex-1 flex flex-col gap-6 border-l-2 border-slate/30 pl-6 relative">
-              <div className="absolute -left-[5px] top-0 w-2 h-2 bg-hazard" />
-              <div className="text-[10px] text-hazard tracking-[0.2em] font-bold">PHASE_03 // EDGE_INTELLIGENCE</div>
-              <BlueprintNode title="SCIKIT-LEARN" desc="Soft-Voting Ensemble" pulse />
-              <BlueprintNode title="AST COMPILER" desc="Python to C Transpiler" />
-              <BlueprintNode title="ESP32 HARDWARE" desc="Zero-Dep C Firmware" pulse />
+          {/* COLUMN 3 */}
+          <div className="bg-canvas-cream flex flex-col justify-between p-8 relative group">
+            <div className="text-[10px] text-hazard tracking-[0.2em] font-bold mb-8">PHASE_03 // EDGE_INTELLIGENCE</div>
+            <div className="flex flex-col gap-12">
+              <div>
+                <h3 className="font-sans text-[24px] font-black uppercase text-ink tracking-tight mb-2 group-hover:text-hazard transition-colors">[ SCIKIT-LEARN ]</h3>
+                <p className="text-[12px] text-slate uppercase leading-tight">Soft-Voting 5-Model Ensemble</p>
+              </div>
+              <div>
+                <h3 className="font-sans text-[24px] font-black uppercase text-ink tracking-tight mb-2 group-hover:text-hazard transition-colors">[ AST COMPILER ]</h3>
+                <p className="text-[12px] text-slate uppercase leading-tight">Python-to-C Zero-Dep Transpiler</p>
+              </div>
+              <div>
+                <h3 className="font-sans text-[24px] font-black uppercase text-ink tracking-tight mb-2 group-hover:text-hazard transition-colors">[ ESP32 HARDWARE ]</h3>
+                <p className="text-[12px] text-slate uppercase leading-tight">PlatformIO INT8 Quantized Firmware</p>
+              </div>
             </div>
-
+            <div className="mt-16 text-[10px] text-slate uppercase border-t border-slate/30 pt-4">TARGET_OUTPUT ///</div>
           </div>
         </motion.div>
 
-        {/* Detailed Specifications */}
-        <motion.div variants={staggerContainer} initial="hidden" animate="show" className="w-full flex flex-col gap-8">
+        {/* SPECIFICATIONS */}
+        <motion.div initial="hidden" animate="show" variants={fadeUp} className="w-full flex flex-col gap-16">
           
-          <motion.div variants={fadeUp} className="border-t-4 border-ink pt-8">
-            <h2 className="text-[20px] md:text-[24px] font-sans font-black tracking-[-0.02em] uppercase text-ink mb-6">
-              [ SPEC_01 ] Data Sanitization & Feature Engineering
-            </h2>
-            <div className="grid grid-cols-1 md:grid-cols-2 gap-8">
-              <div className="bg-white border-2 border-ink p-6">
-                <h3 className="text-[12px] text-hazard font-bold tracking-[0.1em] mb-4 uppercase">/// Pipeline_Execution</h3>
-                <p className="text-[13px] md:text-[14px] text-slate leading-relaxed uppercase mb-4">
-                  Raw telemetry is hostile. The ingestion script aggressively drops redundant alert flags to prevent target-leakage. Broken unicode encodings from edge sensors are patched on the fly.
-                </p>
-                <div className="font-mono text-[10px] md:text-[11px] bg-ink text-canvas-cream p-4 border-l-4 border-hazard">
-                  <span className="text-slate">01</span> DROP: 'Heart Rate Alert', 'Patient Number'<br/>
-                  <span className="text-slate">02</span> PATCH: '\ufffd' -&gt; '°'<br/>
-                  <span className="text-slate">03</span> ENCODE: LabelEncoder(Predicted Disease)
+          {/* SPEC 1 */}
+          <div className="w-full grid grid-cols-1 md:grid-cols-4 gap-8">
+            <div className="md:col-span-1">
+              <div className="text-[10px] text-hazard font-bold tracking-[0.15em] mb-2 uppercase">SPEC_01</div>
+              <h2 className="text-[24px] font-sans font-black tracking-[-0.02em] uppercase text-ink leading-tight">DATA SANITIZATION & FEATURES</h2>
+            </div>
+            <div className="md:col-span-3 bg-ink p-[2px]">
+              <div className="bg-canvas-cream grid grid-cols-1 md:grid-cols-2 h-full gap-[2px] bg-ink">
+                
+                <div className="bg-canvas-cream p-8">
+                  <h3 className="text-[12px] font-bold text-ink tracking-[0.1em] mb-4 uppercase border-b-2 border-ink pb-2">/// Pipeline_Execution</h3>
+                  <p className="text-[13px] text-slate leading-relaxed uppercase mb-6">
+                    Raw telemetry is hostile. The ingestion script aggressively drops redundant alert flags to prevent target-leakage. Broken unicode encodings from edge sensors are patched on the fly.
+                  </p>
+                  <pre className="text-[10px] bg-ink text-canvas-cream p-4 uppercase overflow-hidden leading-loose">
+                    <span className="text-hazard">01</span> DROP: 'Heart Rate Alert'<br/>
+                    <span className="text-hazard">02</span> PATCH: '\ufffd' -&gt; '°'<br/>
+                    <span className="text-hazard">03</span> ENCODE: LabelEncoder(Disease)
+                  </pre>
                 </div>
-              </div>
-              <div className="bg-white border-2 border-ink p-6">
-                <h3 className="text-[12px] text-hazard font-bold tracking-[0.1em] mb-4 uppercase">/// Feature_Synthesis</h3>
-                <p className="text-[13px] md:text-[14px] text-slate leading-relaxed uppercase mb-4">
-                  Missing sensor drops are repaired using a fitted <code className="bg-canvas-cream px-1 text-ink">SimpleImputer(mean)</code>. We inject a non-linear hint directly into the AST: <code className="bg-canvas-cream px-1 text-ink">Risk_Severity</code>. All floats are strictly normalized via <code className="bg-canvas-cream px-1 text-ink">StandardScaler</code>.
-                </p>
-                <div className="font-mono text-[10px] md:text-[11px] bg-ink text-canvas-cream p-4 border-l-4 border-hazard">
-                  Risk_Severity = (HR &gt; 105) + (SpO2 &lt; 94)<br/>
-                  Imputer.fit_transform(continuous_features)<br/>
-                  Scaler.fit_transform(continuous_features)
+                
+                <div className="bg-canvas-cream p-8">
+                  <h3 className="text-[12px] font-bold text-ink tracking-[0.1em] mb-4 uppercase border-b-2 border-ink pb-2">/// Feature_Synthesis</h3>
+                  <p className="text-[13px] text-slate leading-relaxed uppercase mb-6">
+                    Missing sensor drops are repaired using a fitted <code className="bg-ink text-white px-1">SimpleImputer</code>. We inject a non-linear hint directly into the AST: <code className="bg-ink text-white px-1">Risk_Severity</code>. All floats are strictly normalized.
+                  </p>
+                  <pre className="text-[10px] bg-ink text-canvas-cream p-4 uppercase overflow-hidden leading-loose">
+                    <span className="text-hazard">&gt;</span> Risk_Sev = (HR&gt;105) + (SpO2&lt;94)<br/>
+                    <span className="text-hazard">&gt;</span> Imputer.fit_transform(features)<br/>
+                    <span className="text-hazard">&gt;</span> Scaler.fit_transform(features)
+                  </pre>
                 </div>
+
               </div>
             </div>
-          </motion.div>
+          </div>
 
-          <motion.div variants={fadeUp} className="border-t-4 border-ink pt-8">
-            <h2 className="text-[20px] md:text-[24px] font-sans font-black tracking-[-0.02em] uppercase text-ink mb-6">
-              [ SPEC_02 ] Soft-Voting Ensemble & MLOps
-            </h2>
-            <div className="bg-white border-2 border-ink p-6">
-              <p className="text-[13px] md:text-[14px] text-slate leading-relaxed uppercase mb-6 max-w-3xl">
+          {/* SPEC 2 */}
+          <div className="w-full grid grid-cols-1 md:grid-cols-4 gap-8">
+            <div className="md:col-span-1">
+              <div className="text-[10px] text-hazard font-bold tracking-[0.15em] mb-2 uppercase">SPEC_02</div>
+              <h2 className="text-[24px] font-sans font-black tracking-[-0.02em] uppercase text-ink leading-tight">ENSEMBLE & AUTOMATED MLOPS</h2>
+            </div>
+            <div className="md:col-span-3 bg-white border-4 border-ink p-8 relative">
+              <div className="absolute top-0 right-8 w-8 h-full bg-[repeating-linear-gradient(45deg,transparent,transparent_4px,rgba(230,25,25,0.1)_4px,rgba(230,25,25,0.1)_8px)]" />
+              <p className="text-[14px] text-slate leading-relaxed uppercase mb-8 max-w-2xl relative z-10">
                 Single-model inference is clinically insufficient. HeartFlow_OS aggregates probability distributions from 5 disparate algorithms: <strong>KNN, SVM, LogReg, Random Forest, and a Neural Network (MLP)</strong>. 
                 <br/><br/>
-                When a clinician uploads a new batch dataset, the API spawns a background thread. It retrains all 5 models, calculates the weighted F1-Score against a 20% holdout, and checks <code className="bg-canvas-cream px-1 text-ink">registry.json</code>. If the new score degrades, the batch is destroyed. If it improves, the active `.pkl` weights are hot-swapped globally without dropping WebSocket streams.
+                When a clinician uploads a new batch dataset, the API spawns a background thread. It retrains all 5 models, calculates the weighted F1-Score against a 20% holdout, and checks <code className="bg-ink text-white px-1">registry.json</code>. If the new score degrades, the batch is destroyed. If it improves, the active `.pkl` weights are hot-swapped globally without dropping WebSocket streams.
               </p>
-              <div className="flex flex-col md:flex-row gap-4 items-center justify-start text-[12px] font-bold">
-                <div className="bg-ink text-white px-4 py-2">F1 SCORE &gt; ACTIVE_V1</div>
-                <div className="text-ink">━━▶</div>
-                <div className="bg-hazard text-white px-4 py-2 border-2 border-hazard shadow-[4px_4px_0px_0px_rgba(10,10,10,1)]">HOT_SWAP_MODELS()</div>
+              
+              <div className="inline-flex flex-col md:flex-row border-2 border-ink bg-ink gap-[2px] relative z-10">
+                <div className="bg-canvas-cream text-ink px-6 py-4 font-bold text-[12px] flex items-center justify-center">
+                  F1 SCORE &gt; ACTIVE_V1
+                </div>
+                <div className="bg-hazard text-white px-6 py-4 font-bold text-[12px] flex items-center justify-center">
+                  HOT_SWAP_MODELS()
+                </div>
               </div>
             </div>
-          </motion.div>
+          </div>
 
         </motion.div>
-      </motion.div>
+      </div>
     </main>
   );
 }

frontend/src/app/layout.tsx

@@ -56,10 +56,8 @@ export default function RootLayout({ children }: { children: React.ReactNode })
               © 2026 HEARTFLOW_OS. TELEMETRY_SYS.
             </div>
             <div className="flex flex-wrap items-center justify-center gap-4 md:gap-8">
-              <a href="https://huggingface.co/spaces/IDKHowToCodeFR/HEARTFLOW_OS" target="_blank" rel="noopener noreferrer" className="text-[12px] font-bold tracking-[0.05em] uppercase text-ink hover:text-hazard transition-colors">HUGGING_FACE</a>
-              <a href="https://idkhowtocodefr.github.io/HEARTFLOW_OS" target="_blank" rel="noopener noreferrer" className="text-[12px] font-bold tracking-[0.05em] uppercase text-ink hover:text-hazard transition-colors">LIVE_APP</a>
               <Link href="/architecture" className="text-[12px] font-bold tracking-[0.05em] uppercase text-ink hover:text-hazard transition-colors">ARCHITECTURE.MD</Link>
-              <a href={`${process.env.NEXT_PUBLIC_API_URL || "http://localhost:8000"}/docs`} target="_blank" rel="noopener noreferrer" className="text-[12px] font-bold tracking-[0.05em] uppercase text-ink hover:text-hazard transition-colors">API_DOCS</a>
+              <a href={`${process.env.NEXT_PUBLIC_API_URL || "https://idkhowtocodefr-tinyml-backend.hf.space"}/docs`} target="_blank" rel="noopener noreferrer" className="text-[12px] font-bold tracking-[0.05em] uppercase text-ink hover:text-hazard transition-colors">API_DOCS</a>
               <a href="https://github.com/IDKHowToCodeFR/HEARTFLOW_OS" target="_blank" rel="noopener noreferrer" className="text-[12px] font-bold tracking-[0.05em] uppercase text-ink hover:text-hazard transition-colors">GITHUB_REPO</a>
             </div>
           </div>