#3 AI agent to fill Form

 To create an AI agent that can fill out forms using web pages, PDFs, and text sources using Python, you can use:

• Ollama for local LLM inference (e.g., LLaMA3.2).

• LangChain or custom agents for orchestrating form-filling tasks.

• PDF/text parsing via libraries like PyMuPDF, pdfplumber, or docx.

• Web form filling using Selenium.

• Multimodal context support (text + form structure) to extract and align relevant information.

---

🧠 Objective

Create a Python agent that:

1. Reads content from PDFs / web / plain text.

2. Understands and extracts the relevant data.

3. Matches it to form fields using prompt-based reasoning.

4. Uses Ollama with a LLaMA3.2-based model (e.g., lemma) to infer the correct values.

5. Automatically fills a web form using Selenium.

---

📦 Prerequisites

pip install selenium requests flask langchain pdfplumber ollama

You’ll also need:

• A local Ollama instance with LLaMA3/lemma model running:

  ollama run lemma
  

• ChromeDriver installed (for Selenium).

---

🧪 Sample Setup

Here’s a minimal Python setup to create such an agent:

📁 agent.py

import json
import pdfplumber
from selenium import webdriver
from selenium.webdriver.common.by import By
from selenium.webdriver.chrome.options import Options
import time
import requests

OLLAMA_URL = "http://localhost:11434/api/generate"
MODEL_NAME = "lemma"  # or llama3


def extract_text_from_pdf(pdf_path):
    text = ""
    with pdfplumber.open(pdf_path) as pdf:
        for page in pdf.pages:
            text += page.extract_text() + "\n"
    return text


def extract_form_fields(driver):
    fields = {}
    elements = driver.find_elements(By.XPATH, "//input[@type='text' or @type='email' or @type='number' or @type='date']")
    for element in elements:
        try:
            label = element.get_attribute("name") or element.get_attribute("id")
            if label:
                fields[label] = element
        except Exception:
            pass
    return fields


def ask_ollama_to_fill(form_labels, context_text):
    prompt = f"""
You are a helpful assistant. Given the following text:\n\n{context_text}\n\n
Fill out this form with these fields: {list(form_labels)}.
Respond in JSON format with field names and their values.
    """
    response = requests.post(OLLAMA_URL, json={
        "model": MODEL_NAME,
        "prompt": prompt,
        "stream": False
    })
    data = response.json()
    try:
        filled_data = json.loads(data['response'].strip())
        return filled_data
    except:
        print("Could not parse JSON. Response:", data['response'])
        return {}


def fill_web_form(url, context_text):
    options = Options()
    options.add_argument("--headless=new")
    driver = webdriver.Chrome(options=options)
    driver.get(url)
    time.sleep(2)

    form_fields = extract_form_fields(driver)
    field_labels = list(form_fields.keys())
    filled_data = ask_ollama_to_fill(field_labels, context_text)

    for label, value in filled_data.items():
        if label in form_fields:
            form_fields[label].send_keys(str(value))

    print("Form filled. Pausing for review...")
    time.sleep(10)
    driver.quit()


if __name__ == "__main__":
    # Example source: PDF or text
    source_text = extract_text_from_pdf("sample_resume.pdf")
    # Or use plain text: source_text = open("info.txt").read()

    # Example web form URL (local or online)
    form_url = "http://example.com/form"

    fill_web_form(form_url, source_text)

---

📝 Example Prompt Handling

Input PDF Text:

Name: Ramesh Kumar
Email: ramesh@example.com
Phone: 9876543210
DOB: 1995-01-15

Form Labels (HTML fields):

["name", "email", "phone", "dob"]

Ollama Output:

{
  "name": "Ramesh Kumar",
  "email": "ramesh@example.com",
  "phone": "9876543210",
  "dob": "1995-01-15"
}

---

✅ Enhancements

• Add support for textarea, select, and checkbox types.

• Use LangChain Agents with tools like SeleniumTool, PDFLoader, etc.

• Log/Store filled data as JSON for audit.

• Integrate file upload UI using Flask for real-world apps.

---

Would you like me to provide:

• A ready-to-run zip file for this?

• A Flask frontend where users can upload PDF/text and see the form filled?

Let me know how you'd like to expand this.