Upload README.md

README.md

@@ -0,0 +1,164 @@
+---
+title: ExecAssist
+emoji: 📧
+colorFrom: indigo
+colorTo: blue
+sdk: docker
+app_port: 7860
+pinned: false
+license: mit
+tags:
+  - openenv
+  - rl
+  - executive-assistant
+---
+
+# ExecAssist — Executive Assistant Environment
+
+An OpenEnv environment where AI agents learn to manage email and calendar for busy executives.
+
+## Problem Statement
+
+Every executive assistant juggles email, calendars, and scheduling conflicts daily. This environment simulates that exact challenge: read incoming requests, draft professional replies, book meetings, and resolve conflicts intelligently.
+
+**Theme:** #3.2 - World Modeling (Personalized Tasks)
+
+## Tasks
+
+### Task 1: Easy — Simple Meeting Request
+- **Challenge:** Single email with clear calendar availability
+- **Agent must:** Draft polite reply + book meeting in open slot
+- **Score:** 50% email quality + 50% scheduling correctness
+
+### Task 2: Medium — Scheduling Conflict
+- **Challenge:** Requested time is already booked
+- **Agent must:** Identify conflict + propose 2-3 alternatives + explain professionally
+- **Score:** 30% email quality + 40% conflict resolution + 30% scheduling
+
+### Task 3: Hard — Multi-Party Coordination
+- **Challenge:** 3 emails requesting meetings, some overlapping, priority conflicts
+- **Agent must:** Prioritize + reschedule + notify all parties
+- **Score:** 34% email + 33% scheduling + 33% conflict
+
+## Environment Design
+
+### Observation Space
+- **Emails:** Sender, subject, body, priority
+- **Calendar:** Existing meetings, working hours, blocked times
+- **Contacts:** Names, emails, timezones
+
+### Action Space
+```json
+{
+  "email_reply": "Professional response text",
+  "calendar_action": "book | propose_alternatives | reschedule | decline",
+  "meeting_details": {
+    "participants": ["email@company.com"],
+    "start_time": "2026-04-28T14:00:00",
+    "end_time": "2026-04-28T15:00:00",
+    "subject": "Meeting topic",
+    "proposed_alternatives": [...]
+  }
+}
+```
+
+### Reward Functions (Multiple Independent Checks)
+
+**1. Email Quality (0-1)**
+- Politeness markers (thank you, regards)
+- Proper greeting/closing
+- Sufficient detail (20+ words)
+- Professional tone (no negative framing)
+- LLM-as-judge for nuance
+
+**2. Scheduling Correctness (0-1)**
+- No double-booking
+- Within working hours
+- Appropriate duration (15min - 2hrs)
+- All participants included
+
+**3. Conflict Resolution (0-1)**
+- Recognizes conflicts
+- Proposes 2-3 alternatives
+- Explains professionally
+- Prioritizes correctly (for hard task)
+
+**4. Anti-Reward Hacking Penalties**
+- Too short email: -0.3
+- Missing meeting details: -0.4
+- Generic/templated: -0.1
+- Overly long: -0.15
+
+## Baseline Scores
+
+### AI Baseline (Nemotron 3 Super 120B) — Untrained
+| Task | Score |
+|------|-------|
+| Easy | 0.315 |
+| Medium | 0.349 |
+| Hard | 0.346 |
+| **Average** | **0.337** |
+
+*Note: These are pre-training scores. The model struggles with JSON formatting, conflict detection, and professional email composition. Training target: 0.60-0.80*
+
+## Setup & Usage
+
+### Local Development
+
+```bash
+# Clone the repository
+git clone https://huggingface.co/spaces/YourUsername/exec-assist
+cd exec-assist
+
+# Install dependencies
+pip install -r requirements.txt
+
+# Run the server
+uvicorn server.app:app --reload
+
+# Open API docs
+# http://127.0.0.1:8000/docs
+```
+
+### Run Baseline Inference
+
+```bash
+# Set environment variables
+export APIBASEURL=https://openrouter.ai/api/v1
+export MODELNAME=nvidia/nemotron-3-super-120b-a12b:free
+export HFTOKEN=your-api-key
+
+# Run inference
+python inference.py
+```
+
+### Docker
+
+```bash
+docker build -t exec-assist .
+docker run -p 7860:7860 exec-assist
+```
+
+## Training (In Progress — Apr 26)
+
+We will train using TRL + Unsloth:
+1. GRPO trainer setup
+2. Reward shaping
+3. Baseline comparison
+4. Before/after examples
+
+## API Endpoints
+
+| Endpoint | Method | Description |
+|----------|--------|-------------|
+| `/reset?task=easy\|medium\|hard` | POST | Start new episode |
+| `/step` | POST | Submit action, get reward |
+| `/state` | GET | Current state |
+| `/tasks` | GET | List all tasks |
+| `/health` | GET | Health check |
+| `/metadata` | GET | Environment info |
+| `/schema` | GET | Action/observation/state schemas |
+
+## Author
+
+**DevanshuDon** — Built for OpenEnv Hackathon 2026