Internship - LLM-Assisted Reverse Engineering

About Quarkslab

Quarkslab builds cutting-edge cybersecurity solutions used by security-driven companies and institutions around the world. Our QShield product suite focuses on software protection and reverse engineering resistance across desktop, mobile, and embedded platforms.

We’re not in the cloud — we build real software, tested on real systems. If you enjoy diving deep into complex technical environments, automating smart test coverage, and owning quality end-to-end, read on.

Description

Explore how a Large Language Model (LLM) can assist human reverse engineers in understanding compiled binaries (x86/ARM). The goal is to link assembly to semantics, automatically infer behavior, identify key routines, and recognize cryptographic primitives.

During the internship you will work a project with some specific goals and milestones.

1. Reproduce existing research such as “Machine-Language Model for Software Security” (see #bibliography below).

2. Build a full analysis pipeline (binary → disassembly (Ghidra/IDA/Bninja) → pseudo-code → embeddings → LLM-based interpretation.

3. Extend previous work by:

   • Adding an interactive assistant (chat-based RE helper).

   • Evaluating the tool on real binaries (malware, compiled open-source tools).

   • Measuring performance and accuracy of semantic inference.

What you will do

During the internship you will work a project with some specific goals and milestones.

1. Reproduce existing research such as “Machine-Language Model for Software Security” (see #bibliography below).

2. Build a full analysis pipeline (binary → disassembly (Ghidra/IDA/Bninja) → pseudo-code → embeddings → LLM-based interpretation.

3. Extend previous work by:

   • Adding an interactive assistant (chat-based RE helper).

   • Evaluating the tool on real binaries (malware, compiled open-source tools).

   • Measuring performance and accuracy of semantic inference.

Expected Results

• A prototype tool that describes binary behavior using an LLM.

• Quantitative evaluation (accuracy of function descriptions).

• Qualitative evaluation of usefulness for human analysts.

Required Skills

• Programing: Python (intermediate)

• Reverse engineering (intermediate)

• Assembly and binary structures(intermediate)

• Prompt engineering & use of LLM APIs (basic)

Bibliography

• Zhang Chao et al., Machine-Language Models for Software Security

• Shang et. al, BinMetric: A Comprehensive Binary Code Analysis Benchmark for Large Language Models.

• Microsoft Research, CodeXGLUE: A Machine Learning Benchmark Dataset for Code Understanding and Generation.

Assignment

• Get the apksigner app.

• Build a simple pipeline to decompile → analyze → LLM → synthesize.

• In a short document, provide the resulting synthesis and 2 pages explaining how you built the pipeline.