Allan Bendatu

Who I Am

About Me

🤖

AI Architect & Security Researcher

I'm an Informatics student who went deep into the intersection of AI engineering and security research. While most students are writing CRUD apps, I was building multi-agent orchestration systems and discovering real-world vulnerabilities in production AI platforms.

I specialise in zero-cost AI architecture — designing systems that get the power of frontier models without the frontier price tag. I use local models (Ollama) intelligently routed alongside cloud APIs based on task complexity.

🇮🇩 Indonesia 🌏 Remote-first 🤝 Upwork Available 🔍 Open to work

🤖

Agentic AI Systems

Design and build multi-agent pipelines using LangGraph, LangChain, and Ollama. Automated workflows that think, reason, and act autonomously.

🔐

AI Security Research

Audit AI infrastructure for SSRF, prompt injection, and data exfiltration vectors. Hands-on CVE-level research on production AI platforms.

⚡

Automation Engineering

Build Python-powered automation tools: meeting summarisers, report generators, data pipelines — saving teams hours of manual work daily.

Tech Stack

Skills & Technologies

Tools and technologies I use to build production-grade AI systems.

🤖

AI / ML

LangChain LangGraph Ollama OpenAI API Whisper Streamlit

🔐

Security

SSRF Vulnerability Analysis Prompt Injection Pentest Burp Suite

💻

Programming

Python JavaScript Bash / Shell HTML / CSS

🛠

DevOps / Tools

Git / GitHub Docker Linux REST APIs n8n

What I've Built

Featured Projects

Real-world AI systems and security research — fully documented, production-grade work.

Omni-Agent: Zero-Cost AI Orchestrator

⭐ Featured

A production-ready Multi-Agent CLI framework designed to execute security audits, auto-generate documentation, and perform investment analysis — all without expensive API token costs.

Built on a smart routing engine that dispatches tasks to local Ollama models (Llama 3, Mistral) for routine operations and escalates to cloud APIs (OpenAI / Anthropic) only for complex reasoning. This hybrid approach reduces inference costs by up to 90% while maintaining response quality. The system supports pluggable agent modules, making it extensible for any domain.

Python Ollama LangChain CLI View Repository

Security Research: Dify SSRF Vulnerability

⚠ CVE Research Medium-High Severity

In-depth analysis of a Server-Side Request Forgery (SSRF) vulnerability in Dify — a widely-used open-source AI platform with 50k+ GitHub stars.

Discovered that Dify's workflow executor could be manipulated via the base_url parameter to scan internal network topology, bypass IP-allowlist filters, and exfiltrate environment variables from the host. Wrote a full proof-of-concept exploit and responsibly disclosed the findings to the vendor with a detailed reproduction guide.

SSRF Burp Suite Responsible Disclosure Python PoC

AI Meeting Scribe

An autonomous AI agent built from scratch that turns raw meeting recordings into structured, actionable documents in minutes.

The pipeline ingests audio files, transcribes speech using OpenAI Whisper, then feeds the transcript to an LLM agent that produces: a full transcript, an executive summary, a list of key decisions, and categorised action items with owners. Designed for remote teams where async communication is critical — eliminates the need to sit through recorded calls.

Python Whisper LangChain OpenAI View Repository

LangGraph Research Orchestrator

A personal research assistant powered by a LangGraph agentic workflow and a Streamlit interface, designed for iterative, hallucination-free data research.

Unlike a simple chatbot, this system uses a multi-step graph: a planning node decomposes the research question, parallel search nodes retrieve evidence from multiple sources, a verification node cross-checks facts, and a synthesis node produces a cited report. This mirrors the workflow of a junior data analyst — without the risk of AI confabulation. Configurable depth and source preferences via the Streamlit UI.

LangGraph Streamlit Python RAG View Repository

Get In Touch

Let's Work Together

I'm actively looking for remote opportunities on Upwork and LinkedIn. If you need an AI engineer who ships real systems — let's talk.

Ready to build something intelligent?

Fast response guaranteed — usually within 24 hours.

Send Email WhatsApp Me

📧

Email

allanjeremyb@gmail.com

GitHub

@Anxten

Allan Bendatu

Quick Response