import backend import gradio as gr import re, os, random, json, requests, math import pandas as pd from collections import defaultdict from utilit import skills as all_skills from groq import Groq OLLAMA_URL = "http://localhost:11434/api/chat" MODEL = "qwen2.5:1.5b" HIERARCHY = { "Arithmetic Operations": {1: ["whole numbers", "ordering", "number line"], 2: ["integers"], 3: ["decimals"], 4: ["order of operations"]}, "Fractions Decimals Percents": {1: ["equivalent fractions", "conversion"], 2: ["fraction", "decimal"], 3: ["mixed fractions"], 4: ["percent", "ratio", "rate"]}, "Algebra Basics": {1: ["expression", "terms"], 2: ["like terms", "distributive"], 3: ["simplifying"], 4: ["polynomial"]}, "Equations and Inequalities": {1: ["one step", "two step"], 2: ["solving"], 3: ["multi", "more than two"], 4: ["inequality"]}, "Linear Functions": {1: ["pattern", "table"], 2: ["slope"], 3: ["graph"], 4: ["write linear equation", "from situation"]}, "Systems of Equations": {3: ["graphing"], 4: ["substitution", "system"]}, "Geometry": {1: ["angles", "triangle", "polygon"], 2: ["perimeter", "area"], 3: ["volume"], 4: ["surface area", "3d"]}, "Statistics": {1: ["mean", "median", "mode"], 2: ["range"], 3: ["plot", "histogram"], 4: ["scatter", "sampling"]}, "Probability": {1: ["single"], 2: ["venn"], 3: ["two"], 4: ["combinatorics", "counting"]}, "Advanced Algebra": {2: ["factoring"], 3: ["quadratic"], 4: ["quadratic formula"]} } CLUSTERS = { "Arithmetic Operations": ["addition", "subtraction", "multiplication", "division", "order of operations", "computation", "estimation"], "Fractions Decimals Percents": ["fraction", "decimal", "percent", "ratio", "rate"], "Number Theory": ["prime", "factor", "multiple", "divisibility", "gcf", "lcm", "odd", "even"], "Algebra Basics": ["expression", "like terms", "distributive", "polynomial", "simplifying"], "Equations and Inequalities": ["equation", "inequality", "solving", "variable"], "Linear Functions": ["linear", "slope", "y-intercept", "graph", "table"], "Systems of Equations": ["system"], "Geometry": ["angle", "triangle", "polygon", "circle", "perimeter", "area", "volume", "surface area", "prism", "cone", "sphere"], "Statistics": ["mean", "median", "mode", "range", "plot", "histogram", "scatter"], "Probability": ["probability", "venn", "combinatorics", "counting"], "Transformations": ["rotation", "reflection", "translation", "symmetry"], "Coordinate Geometry": ["coordinate", "distance", "midpoint"], "Advanced Algebra": ["quadratic", "factoring", "trinomials"] } SKILL_VIDEOS = { "whole numbers": "https://www.youtube.com/watch?v=m94WTzp14SA", "ordering": "https://www.youtube.com/watch?v=0_fC9Lsn7-8", "number line": "https://www.youtube.com/watch?v=6u3pU9V7NoQ", "integers": "https://www.youtube.com/watch?v=vNguZreVv_g", "decimals": "https://www.youtube.com/watch?v=kgv_UnE_iXQ", "order of operations": "https://www.youtube.com/watch?v=dAgfnK528RA", "equivalent fractions": "https://www.youtube.com/watch?v=qcHHbaKo_7E", "conversion": "https://www.youtube.com/watch?v=do_IbHId2Os", "fraction": "https://www.youtube.com/watch?v=CA9Xp80ZfIs", "decimal": "https://www.youtube.com/watch?v=9VvU6Sow064", "mixed fractions": "https://www.youtube.com/watch?v=1xuf6Z40Y8U", "percent": "https://www.youtube.com/watch?v=JeVSmq1Nrpw", "ratio": "https://www.youtube.com/watch?v=HpdMJaKa9P0", "rate": "https://www.youtube.com/watch?v=LIYfbe7_EaA", "expression": "https://www.youtube.com/watch?v=LpUeZ8T_6mE", "terms": "https://www.youtube.com/watch?v=0AfS8vS4Zks", "like terms": "https://www.youtube.com/watch?v=H_N1_Z9_X7Y", "distributive": "https://www.youtube.com/watch?v=v-6MShC82ow", "simplifying": "https://www.youtube.com/watch?v=3S9S8p_UuU4", "polynomial": "https://www.youtube.com/watch?v=ffLLmV4mZwU", "one step": "https://www.youtube.com/watch?v=L0_K895874A", "two step": "https://www.youtube.com/watch?v=LDIiYKYvvdA", "solving": "https://www.youtube.com/watch?v=bAerID24QJ0", "multi": "https://www.youtube.com/watch?v=f15zA0PhSek", "more than two": "https://www.youtube.com/watch?v=13YFOnR_kio", "inequality": "https://www.youtube.com/watch?v=njpSAtIDZ30", "pattern": "https://www.youtube.com/watch?v=8mGeU_uIs9I", "table": "https://www.youtube.com/watch?v=7u_X6X8tU-A", "slope": "https://www.youtube.com/watch?v=v8VSDg_WQlA", "graph": "https://www.youtube.com/watch?v=f58_XvY9o30", "write linear equation": "https://www.youtube.com/watch?v=eHPTyYbNmxw", "from situation": "https://www.youtube.com/watch?v=m682S8xUuYg", "graphing": "https://www.youtube.com/watch?v=S0_qX4VJhMQ", "substitution": "https://www.youtube.com/watch?v=zsD965-X-Vw", "system": "https://www.youtube.com/watch?v=fNk_zzaMoBA", "angles": "https://www.youtube.com/watch?v=DGKwdHMiqCg", "triangle": "https://www.youtube.com/watch?v=mLeNaZcy-hE", "polygon": "https://www.youtube.com/watch?v=IaoZhhx_I9s", "perimeter": "https://www.youtube.com/watch?v=AAY1bsazcgM", "area": "https://www.youtube.com/watch?v=xCdxURXMdFY", "volume": "https://www.youtube.com/watch?v=qJwecTgce6c", "surface area": "https://www.youtube.com/watch?v=uK5E_T6fX7c", "3d": "https://www.youtube.com/watch?v=ZqzAOZasN9A", "mean": "https://www.youtube.com/watch?v=B1HEzQu68tY", "median": "https://www.youtube.com/watch?v=B1HEzQu68tY", "mode": "https://www.youtube.com/watch?v=B1HEzQu68tY", "range": "https://www.youtube.com/watch?v=B1HEzQu68tY", "plot": "https://www.youtube.com/watch?v=mH_S_zT3qH0", "histogram": "https://www.youtube.com/watch?v=gY-mreidV-4", "scatter": "https://www.youtube.com/watch?v=pGZ37A47nI0", "sampling": "https://www.youtube.com/watch?v=pTuj57uXWlk", "single": "https://www.youtube.com/watch?v=KzfWUEJjG18", "venn": "https://www.youtube.com/watch?v=XdfS2-W8WnU", "two": "https://www.youtube.com/watch?v=Y-XU9rL-pY0", "combinatorics": "https://www.youtube.com/watch?v=Tr6Yp3_78uY", "counting": "https://www.youtube.com/watch?v=O129Sscf6lE", "factoring": "https://www.youtube.com/watch?v=f_Vp_74p_G8", "quadratic": "https://www.youtube.com/watch?v=6QYatS7vV0o", "quadratic formula": "https://www.youtube.com/watch?v=EeVqtpuGgu8" } hierarchy_skills_list = [] for domain_data in HIERARCHY.values(): for skill_group in domain_data.values(): hierarchy_skills_list.extend(skill_group) css = """ body, .gradio-container { font-size: 16px; } """ groq_key = os.getenv("GROQ_KEY") backend.init_db() client = Groq(api_key=groq_key) model_id = "llama-3.3-70b-versatile" session_skills = [] def assign_cluster(skill): skill_lower = str(skill).lower() for cluster, keywords in CLUSTERS.items(): if any(keyword in skill_lower for keyword in keywords): return cluster return "Other" def assign_hierarchy(skill): skill_lower = str(skill).lower() for domain, levels in HIERARCHY.items(): for level, keywords in levels.items(): if any(keyword in skill_lower for keyword in keywords): return domain, level return "Other", -1 def find_most_similar_prediction(current_session_skills): all_logs = backend.get_all_logs() if not all_logs: return [], "Database is empty." student_histories = defaultdict(list) for user_id, skill in all_logs: student_histories[str(user_id)].append(str(skill)) if not student_histories: return [], "No student data available." random_student = random.choice(list(student_histories.keys())) selected_history = student_histories[random_student][-30:] formatted_history = [] for s in selected_history: domain, level = assign_hierarchy(s) formatted_history.append(f"(skill={s}, cluster={assign_cluster(s)}, level={level})") prompt = f"""You are a student knowledge model. Reference Student history: {", ".join(formatted_history)} The current user has expressed difficulty with: {", ".join(current_session_skills)} Available hierarchy skills: {", ".join(hierarchy_skills_list)} Predict the NEXT 3 most likely skills (ranked) for the current user. Reply ONLY with JSON: {{"skills": ["Skill_A", "Skill_B", "Skill_C"]}}""" try: response = requests.post( OLLAMA_URL, json={ "model": MODEL, "messages": [{"role": "user", "content": prompt}], "stream": False, "format": "json" }, timeout=15 ) response.raise_for_status() raw_data = response.json() text_content = raw_data.get("message", {}).get("content", "{}") parsed = json.loads(text_content) preds = parsed.get("skills", []) for i, skill in enumerate(preds, 1): print(f"Rank {i}: {skill}") return preds, f"Random match based on Student ID: {random_student}" except Exception as e: print(f"Detailed LLM Error: {e}") return [], f"LLM processing error: {str(e)}" def get_youtube_id(url): pattern = r'(?:https?://)?(?:www\.)?(?:youtube\.com/(?:watch\?v=|embed/)|youtu\.be/)([a-zA-Z0-9_-]{11})' match = re.search(pattern, url) return match.group(1) if match else None def display_video_and_predict(): preds, info_msg = find_most_similar_prediction(session_skills) print(f"Final Output: {info_msg} Recommendations: {preds}") if not preds: return "

No recommendation available.

", f"### Personalized Path\n{info_msg}" main_skill = preds[0] main_url = SKILL_VIDEOS.get(main_skill, "https://www.youtube.com/watch?v=WUvTyaaNkzM") video_id = get_youtube_id(main_url) video_html = f'''

Main Lesson: {main_skill.title()}

''' other_markdown = f"### Personalized Path\n{info_msg}\n\n#### Other recommendation\n" for skill in preds[1:]: url = SKILL_VIDEOS.get(skill, "#") other_markdown += f"* [{skill.title()}]({url})\n" return video_html, other_markdown def chat_local(message, history): system_content = f"""You are a diagnostic assistant for a maths learning app. Your ONLY job is to identify what the student does not understand in maths. How to behave: 1. Ask the student what topic or concept they are struggling with or don't understand. 2. Ask short clarifying questions if needed to pinpoint the exact difficulty. 3. Once you have enough information about one topic, you MUST respond EXACTLY in this format: "Got it! I'm going to select some exercises to help you with that. [SKILL:skill_name_here]" NEVER say "Got it!" without the [SKILL:] tag. The [SKILL:] tag is mandatory. 4. After saying "Got it!", ask the student if they have any other topic they struggle with and keep the conversation going. 5. Never explain, never teach, never solve anything. Map what the student says to the most relevant skill from this list: {all_skills} NEVER mention the skills list to the student, only include it in the [SKILL:] tag. Rules: - Maximum 2 sentences per response. - Always end with a question. - If the student asks you to explain something, say: "I'm not a tutor! Tell me what you don't understand and I'll find the right exercises for you." """ messages = [{"role": "system", "content": system_content}] for m in history: messages.append({"role": m["role"], "content": m["content"]}) try: response = client.chat.completions.create( model=model_id, messages=messages, max_tokens=300, ) return response.choices[0].message.content except Exception as e: print(f"API Error: {e}") return None def respond(message, chat_history, user_id): global session_skills chat_history.append({"role": "user", "content": message}) bot_message = chat_local(message, chat_history) chat_history.append({"role": "assistant", "content": bot_message or "Sorry, I couldn't generate a response."}) if bot_message and "Got it!" in bot_message: match = re.search(r'\[SKILL:(.*?)\]', bot_message) if match: skill = match.group(1).strip() if skill not in session_skills: session_skills.append(skill) backend.add_skill(user_id, skill) chat_history[-1]["content"] = re.sub(r'\[SKILL:(.*?)\]', f"We'll be working on: {skill}", bot_message).strip() return "", chat_history def handle_login(email, password): user_id = backend.login(email, password) if user_id: return gr.update(visible=False), gr.update(visible=True), "", user_id return gr.update(visible=True), gr.update(visible=False), "## Invalid email or password", None def reset_skills(): global session_skills session_skills = [] with gr.Blocks(theme=gr.themes.Soft(), css=css) as demo: with gr.Column(visible=True) as auth_block: gr.Markdown("# Identification") with gr.Tab("Sign In"): u = gr.Textbox(label="Email") p = gr.Textbox(label="Password", type="password") btn_l = gr.Button("Sign In", variant="primary") out_l = gr.Markdown() with gr.Tab("Create account"): nu = gr.Textbox(label="Username") nemail = gr.Textbox(label="Email") np = gr.Textbox(label="Password", type="password") btn_r = gr.Button("Sign Up") out_r = gr.Textbox(label="Status") btn_r.click(backend.create_user, [nemail, np, nu], out_r) with gr.Column(visible=False) as main_block: gr.Markdown("# AI LEARNING ASSISTANT") user_id_state = gr.State(None) with gr.Row(): with gr.Column(scale=1): chatbot = gr.Chatbot(height=450) msg = gr.Textbox(placeholder="Tell me what you don't understand...", label="Tutor Chat") with gr.Row(): send_btn = gr.Button("Send", variant="primary") clear_btn = gr.Button("Clear") logout_btn = gr.Button("Logout", variant="secondary") with gr.Column(scale=1): video_output = gr.HTML(value="

Video will appear here.

") prediction_output = gr.Markdown() change_video = gr.Button("Get Video Lesson & Recommendation", variant="primary") btn_l.click( fn=handle_login, inputs=[u, p], outputs=[auth_block, main_block, out_l, user_id_state] ) send_btn.click(respond, [msg, chatbot, user_id_state], [msg, chatbot]) msg.submit(respond, [msg, chatbot, user_id_state], [msg, chatbot]) clear_btn.click(lambda: ([], reset_skills()), None, chatbot) change_video.click(fn=display_video_and_predict, outputs=[video_output, prediction_output]) logout_btn.click(lambda: (gr.update(visible=True), gr.update(visible=False), "", None), None, [auth_block, main_block, out_l, user_id_state]) demo.launch()