Jibril-ctrl

Cybersecurity and AI

recent posts

🚀 Agentic Summarizer

” Web App Evolution: Built in Jupyter for rapid iteration, then promoted to a standalone Python application using Gradio. The result is a clean, user-friendly web interface that preserves local execution and privacy, and supports both URL- and file-based summarization. The architecture is intentionally extensible, with additional capabilities implemented privately beyond the open-source baseline. The embedded video below demonstrates the prototype and core workflow, not the deployed web interface. “

Introduction

In today’s world of information overload, the ability to distill long articles into crisp, actionable insights is more valuable than ever.

I built Agentic Summarizer, a lightweight yet powerful AI project that combines Ollama’s local LLMs with LangChain tooling to create a professional summarization agent.

The system runs entirely locally — no API costs, no cloud dependency — and outputs exactly 5 executive-style bullets for any web article or document.

🔑 What Agentic Summarizer Does

  • ✅ Summarizes any single URL into exactly 5 concise, executive bullets.
  • ✅ Handles 3 URLs at once and produces a structured comparison (common themes, differences, unique insights, takeaway).
  • ✅ Saves results in Markdown or PDF format with a signature footer.
  • ✅ Offers an interactive Jupyter Notebook UI with tabs for Single or Multi-URL workflows.
  • ✅ Fully documented with examples, requirements, and walkthroughs.

⚙️ Architecture at a Glance

  • LangChain @tool wrapper → For clean integration of tools like fetch_url (content extraction) and save_note (Markdown export).
  • Ollama (llama3.2:3b) → Fast, local language model for generating structured summaries.
  • ipywidgets + Jupyter Notebook → Interactive tabbed UI for Single URL and Multi-URL batch processing.
  • ReportLab → Optional PDF export with branded footer.

📌 In a perfect world (with budget), I’d also integrate OpenAI GPT-4 or Anthropic Claude for larger context windows and deeper comparison outputs.

📂 Repository Structure

  • Agentic_Summarizer.ipynb → Full interactive notebook
  • src/ → Modularized source code (tools, summarizer, helpers)
  • examples/ → Sample outputs (.md + .pdf)
  • tests/ → Starter test cases
  • requirements.txt → One-line setup
  • Agentic_Summarizer_Walkthrough_Executive.docx → Detailed executive walkthrough

📊 Example Output

Single URL summary:

URL: https://en.wikipedia.org/wiki/Zero_trust_architecture

  1. Zero Trust Architecture is a design strategy assuming no user/device is trusted by default.
  2. Access requires identity verification, device compliance checks, and least-privilege authorization.
  3. Traditional perimeter-based trust (e.g., VPNs) is insufficient for complex modern networks.
  4. Zero Trust enforces mutual authentication and confidence in both user identity and device status.
  5. Data access is governed by Attribute-Based Access Control (ABAC).

3 URLs comparison (excerpt):

  • Common Themes → Zero Trust assumes no inherent trust; all emphasize least-privilege and identity verification.
  • Differences → Defense-in-depth introduces layered security controls, while Zero Trust focuses on per-request verification.
  • Takeaway → Zero Trust and Defense-in-Depth are complementary strategies for modern cybersecurity resilience.

🌟 Why It Matters

This project demonstrates agentic AI design — lightweight tools orchestrated by a local LLM to act autonomously within defined boundaries.

Instead of one-off prompts, the summarizer behaves like a mini agent:

  1. Fetch content
  2. Summarize into structured bullets
  3. Save/export results automatically

It’s fast, private, cost-free, and a real example of applied AI architecture.

📍 Check It Out

👉 GitHub Repo: Agentic Summarizer

👉 Example Walkthrough: https://www.youtube.com/watch?v=yJ3JoZQL5ps

✨ Closing Thoughts

Building this project showed me the power of agentic AI frameworks — even with free local models like Ollama llama3.2:3b, you can deliver production-quality functionality.

Future improvements may include:

  • Integration with OpenAI GPT-4 or Claude for extended context.
  • A web dashboard for non-technical users.
  • Support for additional export formats (PowerPoint, Excel).

✍️ Author: Jibril Anifowoshe — September 2025
Cybersecurity Architect & AI Engineer | $900M Risk Reduction • Zero Trust Design • Advanced Threat Modeling • Incident Response Leadership | Innovating with Agentic AI