Update app.py

Lead capture + email cta fix v2

app.py

@@ -1 +1,214 @@
-Final fix: Lead form + Email CTA colab@anktechsol.com
+import gradio as gr
+import requests
+import os
+
+NL = chr(10)
+
+# Map user-facing modality choices to HF Hub task tags
+TASK_MAP = {
+    "Text in -> text out": ["text-generation", "text2text-generation"],
+    "Images in (docs, UI, scans)": ["image-to-text", "visual-question-answering"],
+    "Audio in (speech)": ["automatic-speech-recognition"],
+    "Structured data (tables, logs)": ["tabular-classification", "tabular-regression"],
+    "Code": ["text-generation"],
+}
+
+LICENSE_FILTER = {
+    "Standard SaaS API is fine": None,
+    "Must stay in a specific cloud region": None,
+    "Strict: prefer on-prem / VPC only": "apache-2.0",
+}
+
+def fetch_top_models(tasks, license_filter=None, top_n=5):
+    seen = set()
+    results = []
+    for task in tasks:
+        try:
+            params = {
+                "pipeline_tag": task,
+                "sort": "downloads",
+                "direction": "-1",
+                "limit": "20",
+            }
+            if license_filter:
+                params["license"] = license_filter
+            resp = requests.get(
+                "https://huggingface.co/api/models",
+                params=params,
+                timeout=15,
+            )
+            if resp.status_code == 200:
+                data = resp.json()
+                for m in data:
+                    mid = m.get("modelId", m.get("id", ""))
+                    if mid and mid not in seen:
+                        seen.add(mid)
+                        downloads = m.get("downloads", 0) or 0
+                        likes = m.get("likes", 0) or 0
+                        results.append((mid, task, downloads, likes))
+                        if len(results) >= top_n * len(tasks):
+                            break
+        except Exception:
+            pass
+    
+    results.sort(key=lambda x: x[2], reverse=True)
+    return results[:top_n]
+
+def format_model_table(models):
+    if not models:
+        return "No models found via HF API. Please try again."
+    
+    header = "| # | Model | Task | Downloads | Likes |" + NL
+    header += "|---|-------|------|-----------|-------|" + NL
+    rows = ""
+    for i, (mid, task, dl, lk) in enumerate(models, 1):
+        dl_fmt = f"{dl:,}" if dl else "N/A"
+        lk_fmt = f"{lk:,}" if lk else "N/A"
+        rows += f"| {i} | [{mid}](https://huggingface.co/{mid}) | `{task}` | {dl_fmt} | {lk_fmt} |" + NL
+    return header + rows
+
+def recommend_model(
+    modality, outputs, domains, data_sensitivity, volume, latency, context_size, customization
+):
+    # Collect HF tasks to query
+    hf_tasks = []
+    for m in modality:
+        hf_tasks.extend(TASK_MAP.get(m, []))
+    
+    # Deduplicate, keep order
+    seen_t = set()
+    unique_tasks = []
+    for t in hf_tasks:
+        if t not in seen_t:
+            seen_t.add(t)
+            unique_tasks.append(t)
+            
+    if not unique_tasks:
+        unique_tasks = ["text-generation"]
+
+    license_filter = LICENSE_FILTER.get(data_sensitivity)
+
+    # Fetch live models
+    live_models = fetch_top_models(unique_tasks, license_filter=license_filter, top_n=5)
+    model_table = format_model_table(live_models)
+
+    # Deployment path
+    if data_sensitivity == "Strict: prefer on-prem / VPC only":
+        deploy_path = "🚀 **Private Self-Hosted** - Run top open-source models above via Ollama or vLLM on your own infra."
+    elif volume == "100,000+":
+        deploy_path = "💰 **Cost-Optimized Scale** - Use provisioned throughput for closed models, or self-host on GPU clusters."
+    else:
+        deploy_path = "⚡ **Serverless API** - Closed models: OpenAI/Anthropic/Google APIs. Open-source: HF Inference Endpoints."
+
+    # Smart tips
+    tips = []
+    if any(d in ["Healthcare", "Finance", "Legal"] for d in domains):
+        tips.append("🔒 **Zero Data Retention (ZDR)** - For regulated domains, enable ZDR on OpenAI/Anthropic/Google or use on-prem.")
+    if latency == "< 500 ms (Instant)":
+        tips.append("⚡ **Pick a smaller distilled variant** - Sort by 'likes' and look for 7B/8B versions of the top model.")
+    if context_size == "32K - 200K tokens (Long)":
+        tips.append("📚 **Add a RAG layer** - Even for long-context models, pair with Pinecone/Weaviate/pgvector for accuracy.")
+    if "Style fine-tuning" in customization:
+        tips.append("🎨 **Fine-tune with QLoRA** - Take the top model from the table above and fine-tune on 1k-5k domain examples.")
+    if "RAG + Tool Calling" in customization:
+        tips.append("🔧 **Enable Tool Calling** - Most top models support function calling / tool use. Check the model card for schema.")
+    
+    if not tips:
+        tips.append("✨ Start with the #1 model in the table above and iterate. A great system prompt gets you 80% of the way.")
+
+    tips_text = (NL + "- ").join(tips)
+    
+    how_to = (
+        "### 📋 Your 3-Step Rollout Guide" + NL +
+        "1. **Sandbox (Week 1):** Clone the top 3 models from the table above. Run 50-100 real queries from your dataset." + NL +
+        "2. **Evaluate (Week 2):** Score on accuracy, latency, and cost per 1K tokens. Eliminate bottom performers." + NL +
+        "3. **Deploy (Week 3):** Integrate the winner via API or self-hosted endpoint. Set up monitoring with LangSmith or Helicone." + NL
+    )
+
+    tasks_label = ", ".join(f"`{t}`" for t in unique_tasks)
+    license_label = f"license: `{license_filter}`" if license_filter else "all licenses"
+
+    summary = (
+        "## 🎉 Your Live AI Model Strategy" + NL + NL +
+        "### 📦 Top Models on HuggingFace Hub Right Now" + NL +
+        f"*Live from HF Hub API | Tasks: {tasks_label} | Filter: {license_label} | Sorted by downloads*" + NL + NL +
+        model_table + NL + NL +
+        "### 🗺️ Best Deployment Path" + NL +
+        deploy_path + NL + NL +
+        "### 💡 Pro-Tips for Your Use Case" + NL +
+        "- " + tips_text + NL + NL +
+        how_to
+    )
+    
+    return summary, gr.update(visible=True)
+
+def handle_lead(name, email, company, infra):
+    if not email:
+        return "Please provide an email to receive your architecture PDF."
+    return f"✅ Thanks {name}! We've queued your custom architecture PDF for {company}. Check your inbox at {email} shortly."
+
+with gr.Blocks(theme=gr.themes.Soft(primary_hue="orange", secondary_hue="slate"), title="AI Model Picker | AnkTechsol") as demo:
+    gr.Markdown("# 🚀 AI Model Selection Wizard")
+    gr.Markdown("**Pick the perfect AI brain for your use case.** This tool queries the HuggingFace Hub live and returns the top trending models for your exact task - no hardcoded lists. By [AnkTechsol](https://anktechsol.com).")
+    
+    with gr.Row():
+        with gr.Column(scale=1):
+            gr.Markdown("### 🔍 Step 1: Describe Your Task")
+            modality = gr.CheckboxGroup(label="What inputs are you working with?", choices=["Text in -> text out", "Images in (docs, UI, scans)", "Audio in (speech)", "Structured data (tables, logs)", "Code"])
+            outputs = gr.CheckboxGroup(label="What output do you need?", choices=["Natural language answer / summary", "Classification / tagging", "Field extraction from text/PDF", "Content generation (copy, emails)", "Scoring / ranking / decision"])
+            domains = gr.CheckboxGroup(label="Your Industry / Domain", choices=["General / consumer", "Ecommerce / SaaS", "Finance", "Healthcare", "Legal", "Internal enterprise knowledge"])
+            
+        with gr.Column(scale=1):
+            gr.Markdown("### ⚙️ Step 2: Set Your Constraints")
+            data_sensitivity = gr.Radio(label="Data Privacy Requirements", choices=["Standard SaaS API is fine", "Must stay in a specific cloud region", "Strict: prefer on-prem / VPC only"], value="Standard SaaS API is fine")
+            volume = gr.Radio(label="Expected Daily Request Volume", choices=["< 1,000", "1,000 - 100,000", "100,000+"], value="< 1,000")
+            latency = gr.Radio(label="Latency Requirement", choices=["< 500 ms (Instant)", "0.5 - 5 s (Standard)", "> 5 s (Batch)"], value="0.5 - 5 s (Standard)")
+            context_size = gr.Radio(label="Max Context Window Needed", choices=["< 4K tokens (Short)", "4K - 32K tokens (Medium)", "32K - 200K tokens (Long)"], value="4K - 32K tokens (Medium)")
+            customization = gr.CheckboxGroup(label="Customization Needs", choices=["Prompt engineering only", "Style fine-tuning", "RAG + Tool Calling"])
+
+    gr.Markdown("### 🎁 What you'll get:")
+    gr.Markdown("- **Top 5 Live Models** from HF Hub matched to your task" + NL + "- **Optimized Deployment Path** (Cloud vs On-Prem)" + NL + "- **3-Week Implementation Roadmap**" + NL + "- **Tailored Cost & Latency Pro-Tips**")
+
+    submit_btn = gr.Button("✨ Fetch Live Models and Generate My Strategy", variant="primary", size="lg")
+    gr.Markdown("> 🕐 *Querying HuggingFace Hub live - may take 5-10 seconds. Please wait after clicking.*")
+    
+    gr.Markdown("---")
+    
+    output_md = gr.Markdown(label="Your Live AI Strategy")
+    
+    with gr.Column(visible=False) as lead_section:
+        gr.Markdown("---")
+        gr.Markdown("### 📧 Get Your Custom Architecture PDF")
+        gr.Markdown("We'll take your results above and generate a detailed 1-page architecture diagram and cost estimate for your stack.")
+        with gr.Row():
+            name = gr.Textbox(label="Name", placeholder="Anuj Karn")
+            email = gr.Textbox(label="Work Email", placeholder="anuj@anktechsol.com")
+        with gr.Row():
+            company = gr.Textbox(label="Company / Project", placeholder="AnkTechsol")
+            infra = gr.Dropdown(label="Current Infra", choices=["AWS", "GCP", "Azure", "On-Prem", "Other/None"], value="AWS")
+        
+        lead_btn = gr.Button("📥 Send Me My Architecture PDF", variant="primary")
+        lead_status = gr.Markdown("")
+        
+        gr.Markdown("### ✉️ Ready to build?")
+        gr.Markdown("Contact us at [**colab@anktechsol.com**](mailto:colab@anktechsol.com) to review these results with our architecture team and map your production roadmap.")
+
+    submit_btn.click(
+        fn=recommend_model, 
+        inputs=[modality, outputs, domains, data_sensitivity, volume, latency, context_size, customization], 
+        outputs=[output_md, lead_section]
+    )
+    
+    lead_btn.click(
+        fn=handle_lead,
+        inputs=[name, email, company, infra],
+        outputs=lead_status
+    )
+
+    gr.Markdown("---")
+    gr.Markdown("### 🛠️ Why Teams Trust AnkTechsol")
+    gr.Markdown("We help startups and enterprises go from AI confusion to production-grade AI systems. Fine-tuning, RAG pipelines, GPU inferencing as a service - we deliver.")
+    gr.Markdown("**[Visit anktechsol.com](https://anktechsol.com) | [LinkedIn](https://www.linkedin.com/company/anktechsol)**")
+
+if __name__ == "__main__":
+    demo.launch()