Redesigned the ui of assigning task

.gitignore

@@ -1,5 +1,8 @@
 # See https://help.github.com/articles/ignoring-files/ for more about ignoring files.
 
+# env file
+.env.local
+
 # dependencies
 /node_modules
 /.pnp
@@ -32,7 +35,24 @@ yarn-error.log*
 
 # vercel
 .vercel
+.vscode
 
 # typescript
 *.tsbuildinfo
 next-env.d.ts
+
+# python
+__pycache__/
+*.py[cod]
+ML/env/
+.env
+venv/
+env/
+ML/my_setfit_model/
+ML/skill_matcher_model/
+*.safetensors
+
+# Retrained ML Models
+ML/my_setfit_model_critical/
+
+/supabase

.vscode/settings.json

@@ -0,0 +1,3 @@
+{
+    "liveServer.settings.port": 5501
+}

ML/README_ML.md

@@ -0,0 +1,124 @@
+ML Folder — TaskFlow
+=====================
+
+This document explains the purpose, components, and runtime behavior of the legacy `ML/` folder. It describes the Python FastAPI service and the supporting models used for priority prediction, assignment, urgency scoring, wellness, and bottleneck detection.
+
+Files
+-----
+- `main.py` — FastAPI server that exposes endpoints used by the UI or experiments:
+  - `POST /analyze_task` — returns priority, urgency, detected skills, and suggested assignees for a given task payload.
+  - `POST /analyze_wellness` — returns a wellness score and status for a user's workload inputs.
+  - `POST /analyze_bottlenecks` — analyzes a list of tasks to detect process bottlenecks (WIP breaches, overdue, aging WIP) and computes an overall health score.
+  - `GET /` — health check.
+
+- `models.py` — core model classes and request schemas:
+  - `get_device()` — checks GPU availability, falls back to CPU on failure.
+  - `TaskPriorityModel` — SetFit-based classifier for predicting task priority (Low/Medium/High). Loads model from `my_setfit_model`.
+  - `TaskAssigner` — semantic skill-matching and candidate ranking using sentence-transformers embeddings, combined with wellness scores and simple role-based multipliers.
+  - `UrgencyModel` — deterministic urgency scoring by priority, due-date proximity, staleness, and status multipliers. Also provides a `label()` helper to map numeric score to descriptive label.
+  - Pydantic request models: `FullTaskRequest`, `WellnessRequest`, `BottleneckRequest` + `BottleneckTask`.
+
+- `wellness_model.py` — deterministic wellness calculator:
+  - `WellnessModel.calculate(active_tasks, high_priority_count, critical_urgency_count)` returns 0–100 wellness score.
+  - `get_status(score)` returns human label ("Healthy Balance", "Nearing Capacity", "Overworked", "Burnout Risk").
+  - Tunable constants in the class define comfortable load and penalty weights.
+
+- `requirements.txt` — Python dependencies used to run this service (SetFit, sentence-transformers, FastAPI, torch, etc.).
+
+- `data.txt` — sample or archived data (inspect for training notes).
+
+How the pieces fit
+------------------
+1. `main.py` initializes models on startup:
+   - Loads `TaskPriorityModel` from `MODEL_PATH` (SetFit model directory).
+   - Loads a `SentenceTransformer` model from `SKILL_MODEL_PATH` for semantic skill embeddings.
+   - Instantiates `TaskAssigner`, `UrgencyModel`, and `WellnessModel`.
+   - `get_device()` decides whether to use `cuda` or `cpu`.
+
+2. `POST /analyze_task` flow:
+   - Reads a `FullTaskRequest` including `description`, `status`, `days_until_due`, `days_since_update`, and `candidates` (list of users and skills + wellness data).
+   - Calls `priority_ai.predict(task.description)` to get predicted priority and confidence (SetFit).
+   - Uses `assigner_ai.find_best_match(...)` which:
+     - Encodes candidate skill lists and the task description into embeddings.
+     - Computes cosine similarity to rank skills and extract required skills.
+     - Scores each candidate by skill overlap and their wellness score (if provided), applies role multipliers, produces `combined_ranking_score`.
+   - Uses `urgency_ai.predict(...)` to compute a numeric urgency score and label.
+   - Returns analysis + top suggested assignees (up to 5).
+
+3. `POST /analyze_wellness` and `wellness_model.py`:
+   - `WellnessModel.calculate` is deterministic and uses configurable penalties and a comfortable load constant.
+   - Provides a reliable, explainable wellness score for a user given active/high/critical counts.
+
+4. `POST /analyze_bottlenecks`:
+   - This endpoint is rule-based and does not rely on heavy ML models.
+   - Computes overdue tasks per project, WIP limit breaches for key statuses, and aging WIP (stale tasks). It returns a list of detected bottlenecks, a health score and summary. It is useful as a companion to the main assignment/priority flows.
+
+Design notes & runtime behavior
+-------------------------------
+- Hybrid design: the folder runs an optional ML API server (Python) that uses SetFit and sentence-transformers when available. However the main Next.js app implements TypeScript fallbacks (`lib/ml-engine.ts` and `lib/ml-transformers.ts`) that reproduce similar heuristics so the web app does not strictly depend on this Python service.
+
+- Device detection: `get_device()` attempts a GPU test and falls back to CPU to avoid startup errors on machines with misconfigured CUDA.
+
+- Role weighting in assignment: the assigner intentionally boosts Members (1.1x) and reduces Manager/Admin combined scores to avoid assigning cross-role responsibilities by default.
+
+- Explainability: Most scoring logic in `models.py` and `wellness_model.py` is explicit and deterministic, making it easy to tweak penalties and thresholds for team preferences.
+
+Run locally (development)
+-------------------------
+1. Create a Python virtual environment and install deps:
+
+```bash
+python -m venv .venv
+# PowerShell
+.\.venv\Scripts\Activate.ps1
+# or Bash
+# source .venv/bin/activate
+pip install -r ML/requirements.txt
+```
+
+2. Prepare models and assets:
+- Place a SetFit model export under `ML/my_setfit_model/` (or update `MODEL_PATH` in `main.py`).
+- Place a sentence-transformers model under `ML/skill_matcher_model/` (or update `SKILL_MODEL_PATH`).
+
+If you do not have models locally, most endpoints that rely on them will fail; these modules are often used for offline experiments or higher-quality enrichment.
+
+3. Start the FastAPI server (dev):
+
+```bash
+cd ML
+python main.py
+# or
+uvicorn main:app --host 127.0.0.1 --port 8000 --reload
+```
+
+Endpoint examples
+-----------------
+- Analyze a task (JSON payload):
+
+POST http://127.0.0.1:8000/analyze_task
+Content-Type: application/json
+
+{
+  "description": "Improve DB query performance for invoice search",
+  "status": "To Do",
+  "days_until_due": 2,
+  "days_since_update": 5,
+  "candidates": [
+    {"id":"u1","name":"Alex","role":"Member","skills":["sql","postgres"],"wellness_data":{"active_tasks":3,"high_priority_count":1,"critical_urgency_count":0}},
+    {"id":"u2","name":"Jamie","role":"Member","skills":["frontend","react"],"wellness_data":{"active_tasks":1,"high_priority_count":0,"critical_urgency_count":0}}
+  ]
+}
+
+Response includes `analysis` (priority, confidence, urgency) and `suggested_assignees`.
+
+Maintenance notes
+-----------------
+- The Python `ML/` service is useful for experiments and can be deployed separately for heavier inference.
+- The main Next.js app contains `lib/ml-transformers.ts` and `lib/ml-engine.ts` which implement TypeScript-based heuristics and optional HF calls; prefer updating those for runtime app behavior.
+- The SetFit model format should be exported from the SetFit training workflow and placed under `my_setfit_model`.
+
+Questions or follow-ups
+----------------------
+- Want a sequence diagram showing request -> Python ML server -> models -> response? I can generate it.
+- Want me to add a short `health-check` script or a minimal `dockerfile` for this service?
+