AI Semantic Analyzer — LLM-Powered CSV/Excel Text Classification & Sentiment Analysis Pipeline 🧠📊

AI-powered semantic text analysis for CSV & Excel files. Batch sentiment analysis, text classification, survey response processing, and customer feedback extraction — 5,000+ rows per hour.

Quick Start • Use Cases • Architecture • Who Is This For

---

---

Quick Start

# Clone and install dependencies
cd ai-semantic-analyzer
pip install -r requirements.txt

# Set LLM API key (choose one)
export OPENAI_API_KEY="sk-..."
# or
export DEEPSEEK_API_KEY="..."

# Run analysis
python analyzer.py --input data.csv --output analyzed.csv --batch-size=50

Output: analyzed.csv with additional columns:

• sentiment (positive/negative/neutral)
• sentiment_confidence (0-1)
• category (your custom classification)
• category_confidence (0-1)

---

🌪 The Problem

You have 10,000 customer reviews, survey responses, support tickets, or lead descriptions. Spreadsheet data with unstructured text.

• Manual review? Takes weeks. Expensive. Biased.
• Classic keyword search? Misses sarcasm ("Great service... NOT"), context, and implicit intent.
• Generic AI tools? Hallucinate or fail to output structured data you can actually use in Excel.
• No-code tools? Limited to 500 rows, hit paywalls fast.

⚡ The Solution

AI Semantic Analyzer is an industrial-grade NLP pipeline for CSV and Excel files. It uses Large Language Models (DeepSeek V3, GPT-4, Claude, Ollama) to analyze text data row-by-row — extracting sentiment, classifying content, detecting sarcasm, scoring leads, and more.

Feature	🐢 Manual / Legacy	🚀 AI Semantic Analyzer
Throughput	100 rows / hour	5,000+ rows / hour
Cost	$$$ (Human labor)	<$0.10 per 1k rows (DeepSeek/Gemma)
Analysis Depth	Surface level	Deep semantic & psychological profiling
Output Format	Vague notes	Strict JSON / Structured CSV Columns
LLM Support	N/A	DeepSeek, GPT-4, Claude, Ollama (local)

---

👥 Who Is This For

Role	Use Case
Data Analysts	Batch process survey results, NPS responses, open-ended feedback
Customer Success Teams	Classify support tickets by urgency & topic
Sales & Marketing	Score leads from free-text inquiries, analyze competitor reviews
Researchers	Qualitative coding at scale — sentiment, theme extraction, categorization
E-Commerce Ops	Process product reviews, detect sarcasm, extract feature requests

---

Security & Privacy

⚠️ Important:

• Input files are sent to external LLM APIs (OpenAI, DeepSeek, etc.)
• Do NOT process sensitive/PII data (passwords, personal IDs, confidential info)
• API keys are stored in environment variables — never commit .env
• Processed data may be logged by LLM provider — check provider's data policy
• For on-premise processing, use local LLM (Ollama) with --model llama2
• Output files inherit sensitivity of input — handle accordingly

---

🏗 Architecture

The system uses a Scatter-Gather pattern for parallel AI processing — maximum throughput without rate-limit bottlenecks.

graph LR
  A[📄 Raw CSV/Excel] --> B{⚡ ThreadPool Orchestrator}
  B --> C[🤖 Agent 1]
  B --> D[🤖 Agent 2]
  B --> E[🤖 Agent 3]
  B --> F[🤖 Agent 4]
  C & D & E & F --> G[🧠 LLM Inference (DeepSeek/GPT/Claude/Ollama)]
  G --> H[📦 JSON Extraction]
  H --> I[📊 Structured CSV Output]

Loading

🔥 Key Features

• 🚀 Multi-Threaded Engine — Python ThreadPoolExecutor saturates API limits safely, processes thousands of rows in minutes.
• 🎯 Strict JSON Output — Prompt engineering enforces 100% machine-readable results. No "Here is the analysis" fluff — just structured data.
• 🛡 Row-Level Context Isolation — Every row analyzed independently. No data leakage between customers or samples.
• 🔌 LLM Agnostic — Optimized for DeepSeek V3 (best cost/performance), compatible with OpenAI GPT-4, Anthropic Claude, or local Ollama/Llama models.
• 📊 Excel-Native Output — Results come as CSV with structured columns ready for pivot tables, BI tools, or further analysis.

---

🚀 Quick Start

1. Clone & Install

cd ai-semantic-analyzer
pip install pandas openai openpyxl

2. Configure API Key

Set your LLM API key as an environment variable:

export DEEPSEEK_API_KEY="your-key-here"
# Or for OpenAI: export OPENAI_API_KEY="sk-..."

3. Run Semantic Analysis

python analyzer.py "reviews.csv" "Analyze sentiment (Positive/Negative/Neutral), detect Sarcasm (true/false), and extract Main_Complaint."

---

💡 Use Cases

A. E-Commerce Review Analysis

Input:

"Oh fantastic, another update that breaks the login button. Just what I needed on a Monday."

Task: "Extract sentiment, check for sarcasm, identify broken feature."

Output:

{
 "sentiment": "Negative",
 "sarcasm": true,
 "broken_feature": "Login Button",
 "urgency": "High"
}

B. Lead Scoring & Sales Intelligence

Input:

"We are looking to replace our enterprise CRM for 500 seats next quarter. Budget is flexible."

Task: "Identify intent, company size, budget sensitivity, lead score 1-100."

Output:

{
 "intent": "Purchase",
 "company_size": "Enterprise (500 seats)",
 "budget_sensitivity": "Low",
 "lead_score": 95
}

C. Support Ticket Classification

Input:

"Your payment gateway keeps declining my card. I've tried 3 different cards. FIX THIS NOW."

Task: "Classify ticket topic, urgency level, and sentiment."

Output:

{
 "topic": "Payment Gateway",
 "urgency": "Critical",
 "sentiment": "Angry",
 "needs_escalation": true
}

---

🛠 Roadmap

• Web GUI — Drag-and-drop interface for non-technical users
• Ollama Native Mode — Fully offline local LLM processing
• Smart Batching — Dynamic batch sizing to minimize token costs
• Multi-File Processing — Batch process entire folders of CSV/Excel files
• Streaming Mode — Real-time analysis for live data pipelines

---

Efficient. Brutal. Effective.

⭐ Star this repo if you find it useful — it helps others discover the tool!

License

MIT — see LICENSE

Author

Agent Hermes — Lisa Carter
GitHub @Maatq1544