🎯 Enterprise AI Lead Scoring & Qualification System

Reference Architecture: An automated lead processing pipeline that combines technical validation with LLM-based intent analysis (Llama 3.3) to prioritize sales efforts.

---

💡 The Business Problem: Sales Team Overload

Sales departments often waste their time on low-quality leads or spam. This system acts as an Automated Gatekeeper, ensuring that only high-intent, validated leads reach the CRM (HubSpot), while technical data is cleansed and enriched in the background.

---

🛠️ System Architecture

The workflow follows a 3-stage qualification process:

1. Technical Validation: Cleans phone numbers, validates email domains, and checks if the lead is using a corporate or free (Gmail/WP) email.
2. AI Intent Analysis: Uses Llama 3.3 (via Groq) to analyze the message content, identify the industry, and assign an intent score (HOT/WARM/COLD).
3. CRM Orchestration: Aggregates scores and automatically routes the lead to HubSpot and Google Sheets with appropriate prioritization labels.

graph LR
    Webhook[Inbound Lead] --> Tech[Technical Validation]
    Tech --> AI[Llama 3.3 Analysis]
    AI --> Logic[Final Scoring Logic]
    Logic --> HubSpot[HubSpot CRM]
    Logic --> Sheets[Google Sheets Log]

Loading

---

🧠 Engineering Challenges & Solutions

1. Data Sanitization & Phone Normalization

Problem: Inbound leads often provide phone numbers in inconsistent formats (with prefixes, spaces, or dashes), which causes synchronization errors in CRM systems like HubSpot.
Solution: Built a custom JavaScript processing engine that uses Regex to strip non-numeric characters and normalize international prefixes. This ensures a clean, 9-digit format that is 100% compatible with standard CRM database schemas.

2. Intent Detection via Llama 3.3 (The "Context" Problem)

Problem: Simple keyword-based filters often misclassify leads (e.g., a user asking "how to cancel" vs "how to buy a license").
Solution: Implemented a sophisticated system prompt for Llama 3.3 (70B) that performs qualitative analysis. The model evaluates the "psychological intent" behind the message and assigns an AI-Score based on buying signals, urgency, and professional depth.

3. CRM Hygiene & Lead Prioritization

Problem: Sales teams are often overwhelmed by "noise" (low-quality leads and spam), leading to missed opportunities with high-value prospects.
Solution: Developed a weighted scoring matrix. Only leads that achieve a Total Score > 60 are tagged as 🔥 HOT LEAD. The system automatically updates the "Job Title" field in HubSpot with the Lead Status and detected Industry, allowing sales reps to filter and react to the best opportunities in minutes.

---

📂 Repository Structure

• workflow_sanitized.json: The complete n8n workflow file, ready for import. All sensitive IDs and credentials have been replaced with placeholders.
• scoring_logic.js: A standalone JavaScript file containing the core sanitization and score aggregation logic used in the Node.js nodes.

---

🤝 Transparency: Human vs. AI Role

• My Role (Architect): Defined the lead qualification criteria (Scorecard), designed the multi-stage logic flow, and managed the HubSpot CRM data mapping.
• AI Role: Assisted in fine-tuning the Llama 3.3 system prompt for better classification accuracy and provided boilerplate code for the JavaScript regex normalization.

---

🚀 Deployment Requirements

• Orchestrator: n8n (Self-hosted or Cloud)
• Inference Engine: Groq Cloud API (Llama-3.3-70b-versatile)
• CRM Integration: HubSpot (App Token with Contacts Scopes)
• Storage: Google Sheets API