import gradio as gr import threading import os import pickle from flask import Flask, request, jsonify from environment import DocumentClassificationEnv from baseline_inference import run_task, load_or_train from agent import TicketAgent print("Pre-training models...") models = {} for d in ["easy", "medium", "hard"]: models[d] = load_or_train(d) print(f" {d} model ready!") print("All models ready!") agent = TicketAgent(model_dir=".") print("Agent ready!") CATEGORIES = { "easy": ["General","Billing","Support","Technical","HR"], "medium": ["General","Billing","Support","Technical","HR","Legal","Sales","Marketing","Operations","Complaints"], "hard": [f"Cat_{i}" for i in range(22)] } api = Flask(__name__) api_envs = {} @api.route("/api/reset", methods=["POST"]) def api_reset(): data = request.json or {} difficulty = data.get("difficulty", "easy") seed = data.get("seed", 42) env = DocumentClassificationEnv(task_difficulty=difficulty, seed=seed) obs, _ = env.reset(seed=seed) api_envs["current"] = env return jsonify({"observation": {k: v.tolist() if hasattr(v, 'tolist') else v for k, v in obs.items()}}) @api.route("/api/step", methods=["POST"]) def api_step(): data = request.json or {} action = data.get("action", 0) env = api_envs.get("current") if env is None: return jsonify({"error": "Call /api/reset first"}), 400 obs, reward, done, _, info = env.step(int(action)) return jsonify({ "observation": {k: v.tolist() if hasattr(v, 'tolist') else v for k, v in obs.items()}, "reward": reward, "done": done, "info": info }) @api.route("/api/run", methods=["POST"]) def api_run(): data = request.json or {} difficulty = data.get("difficulty", "easy") score, t = run_task(difficulty) return jsonify({"difficulty": difficulty, "score": score, "time": t}) @api.route("/api/agent/process", methods=["POST"]) def api_agent_process(): data = request.json or {} text = data.get("text", "") difficulty = data.get("difficulty", "easy") ticket_id = data.get("ticket_id", None) if not text: return jsonify({"error": "text field required"}), 400 result = agent.process_ticket(text, difficulty=difficulty, ticket_id=ticket_id) return jsonify(result) @api.route("/api/agent/classify", methods=["POST"]) def api_agent_classify(): data = request.json or {} text = data.get("text", "") difficulty = data.get("difficulty", "easy") category, scores = agent.classify(text, difficulty) priority = agent.get_priority(text) return jsonify({"category": category, "priority": priority, "scores": scores}) def run_flask(): api.run(host="0.0.0.0", port=7861, debug=False) env_state = {"env": None, "difficulty": "easy"} def create_env(difficulty): e = DocumentClassificationEnv(task_difficulty=difficulty, seed=42) obs, _ = e.reset() env_state["env"] = e env_state["difficulty"] = difficulty model = models[difficulty] cats = CATEGORIES[difficulty] proba = model.predict_proba([obs["content"]])[0] top3 = sorted(enumerate(proba), key=lambda x: -x[1])[:3] explain = "\n".join([f" {cats[i] if i < len(cats) else f'Cat_{i}'}: {p*100:.1f}%" for i,p in top3]) pred_idx = int(model.predict([obs["content"]])[0]) pred_cat = cats[pred_idx] if pred_idx < len(cats) else f"Cat_{pred_idx}" return (f"āœ… Environment created! Difficulty: {difficulty}", obs["content"], f"šŸ¤– Model predicts: **{pred_cat}**\n\nTop 3 predictions:\n{explain}", "") def classify_doc(category_idx): e = env_state["env"] if e is None: return "Create environment first!", "", "", "" obs, reward, done, _, info = e.step(int(category_idx)) result = f"Reward: {reward:.2f} | Correct: {info.get('is_correct', False)}" if done: acc = info.get('episode_accuracy', 0) result += f"\nšŸ Episode done! Final Accuracy: {acc:.2%}" return result, "Episode complete — create new environment", "", str(info) difficulty = env_state["difficulty"] model = models[difficulty] cats = CATEGORIES[difficulty] proba = model.predict_proba([obs["content"]])[0] top3 = sorted(enumerate(proba), key=lambda x: -x[1])[:3] explain = "\n".join([f" {cats[i] if i < len(cats) else f'Cat_{i}'}: {p*100:.1f}%" for i,p in top3]) pred_idx = int(model.predict([obs["content"]])[0]) pred_cat = cats[pred_idx] if pred_idx < len(cats) else f"Cat_{pred_idx}" return (result, obs["content"], f"šŸ¤– Model predicts: **{pred_cat}**\n\nTop 3 predictions:\n{explain}", str(info)) def run_baseline_all(): rows = [] for d in ["easy", "medium", "hard"]: score, t = run_task(d) rows.append([d.upper(), f"{score:.4f}", f"{score*100:.1f}%", f"{t:.1f}s"]) return rows def process_ticket_ui(ticket_text, difficulty): if not ticket_text.strip(): return "āŒ Please enter ticket text!", "", "", "", "", "" result = agent.process_ticket(ticket_text, difficulty=difficulty) category_out = f"šŸ·ļø {result['category']} (confidence: {result['confidence']*100:.1f}%)" priority_out = f"🚨 {result['priority'].upper()}" dept_out = f"šŸ¢ {result['department']}\nšŸ“§ {result['email']}\nā±ļø SLA: {result['sla']}" top3_out = "\n".join([f" {c}: {p*100:.1f}%" for c,p in result['top3']]) ref_out = f"šŸŽ« {result['ref_id']} | ā° {result['timestamp'][:19]}" reply_out = result['reply'] return category_out, priority_out, dept_out, top3_out, ref_out, reply_out SAMPLE_TICKETS = { "Billing complaint": "My invoice shows an incorrect amount. I was charged $150 but should have been charged $75. Please review and correct this immediately.", "Bug report": "The application crashes whenever I try to upload a file larger than 10MB. This is a critical issue affecting my workflow.", "HR query": "I have a question about the company's maternity leave policy. How many weeks of paid leave am I entitled to?", "Chat message": "Hey, I can't log into my account. The password reset link isn't working either. Please help ASAP!", "Email ticket": "Dear Support, I would like to inquire about upgrading my current subscription plan to include additional users.", } def load_sample(sample_name): return SAMPLE_TICKETS.get(sample_name, "") with gr.Blocks(title="Document Classification OpenEnv") as demo: gr.Markdown("# šŸ“„ Document Classification OpenEnv") gr.Markdown("Real-world customer support ticket routing environment for RL agent training.") with gr.Tab("šŸ¤– Agent Demo"): gr.Markdown("### Full Automation — ML Classification + Rule-based Priority + Auto-Reply") with gr.Row(): with gr.Column(scale=2): sample_dd = gr.Dropdown( choices=list(SAMPLE_TICKETS.keys()), label="šŸ“‹ Load Sample Ticket", value=None ) ticket_input = gr.Textbox( label="āœļø Ticket / Message Text", lines=6, placeholder="Paste email, chat message, bug report, HR query..." ) diff_agent = gr.Radio(["easy","medium","hard"], value="medium", label="Model Difficulty") btn_process = gr.Button("šŸš€ Process Ticket", variant="primary") with gr.Column(scale=2): ref_out_box = gr.Textbox(label="šŸŽ« Reference ID & Timestamp") cat_out_box = gr.Textbox(label="šŸ·ļø Category & Confidence") pri_out_box = gr.Textbox(label="🚨 Priority") dept_out_box = gr.Textbox(label="šŸ¢ Routed To", lines=3) top3_out_box = gr.Textbox(label="šŸ“Š Top 3 Predictions", lines=3) reply_out_box = gr.Textbox(label="šŸ“§ Auto-Generated Reply", lines=12) sample_dd.change(load_sample, inputs=sample_dd, outputs=ticket_input) btn_process.click( process_ticket_ui, inputs=[ticket_input, diff_agent], outputs=[cat_out_box, pri_out_box, dept_out_box, top3_out_box, ref_out_box, reply_out_box] ) with gr.Tab("šŸŽ® Interactive Demo"): diff = gr.Radio(["easy","medium","hard"], value="easy", label="Difficulty") btn_create = gr.Button("Create Environment", variant="primary") status = gr.Textbox(label="Status") with gr.Row(): doc_content = gr.Textbox(label="šŸ“ Document Content", lines=6) explain_box = gr.Textbox(label="🧠 Explainability — Model Reasoning", lines=6) category = gr.Number(label="Category Index (easy: 0-4, medium: 0-9, hard: 0-21)", value=0) btn_classify = gr.Button("Classify Document") result = gr.Textbox(label="Result") info_box = gr.Textbox(label="Info") btn_create.click(create_env, inputs=diff, outputs=[status, doc_content, explain_box, result]) btn_classify.click(classify_doc, inputs=category, outputs=[result, doc_content, explain_box, info_box]) with gr.Tab("šŸ“Š Baseline Evaluation"): gr.Markdown("TF-IDF + Logistic Regression baseline — models pre-trained at startup for instant results.") btn_eval = gr.Button("Run All Tasks", variant="primary") score_table = gr.Dataframe(headers=["Task","Score","Accuracy","Time"], label="Results") btn_eval.click(run_baseline_all, outputs=score_table) with gr.Tab("šŸ“‹ Environment Info"): gr.Markdown(""" ## Environment Design **Task**: Classify customer support documents into correct departments. | Difficulty | Categories | Episodes | Reward | |------------|-----------|----------|--------| | Easy | 5 | 20 | +1.0 correct, -0.4 wrong | | Medium | 10 | 30 | +1.0 correct, -0.4 wrong | | Hard | 22 | 50 | +1.0 correct, -0.4 wrong | ## Agent Pipeline `Ticket Input` → `ML Classify` → `Rule Priority` → `Department Route` → `Auto Reply` ## API Endpoints (port 7861) - `POST /api/reset` — `{"difficulty": "easy", "seed": 42}` - `POST /api/step` — `{"action": 0}` - `POST /api/agent/process` — `{"text": "my invoice is wrong", "difficulty": "medium"}` - `POST /api/agent/classify` — `{"text": "urgent bug", "difficulty": "easy"}` """) if __name__ == "__main__": t = threading.Thread(target=run_flask, daemon=True) t.start() demo.launch(server_name="0.0.0.0", server_port=7860)