Executable File Analyzer - Interactive Architecture

This section details each component of the architecture. The system is designed for efficiency, using parallel processing for speed and streaming for a real-time user experience.

User's Browser (Client)

Technology: HTML, JavaScript, Tailwind CSS.

Responsibility: Renders the UI and handles the initial file upload. It uses the `fetch` API to make a `POST` request and then processes the response as a real-time `text/event-stream`, updating the display dynamically as results for each function and the final verdict arrive.

Cloud Run Service

Technology: Docker, Python, Flask, Gunicorn.

Responsibility: The core serverless backend. Its **Streaming Analyzer** manages a **ThreadPoolExecutor**, which assigns each code chunk to a separate worker thread. These threads make concurrent API calls, drastically speeding up the analysis phase before results are streamed back.

Generative AI API

Technology: Managed cloud service for AI models.

Responsibility: Provides access to multiple foundation models. It receives parallel requests for individual function analysis and one final request for the overall report, returning structured JSON and text as responses.

The following steps outline the journey of a file from preparation to final analysis report.

1

Preparation: A user first decompiles a binary executable (.exe, .dll) into C/C# source code using a tool like IDA Pro or dnSpy.
2

File Upload: The user uploads the resulting decompiled source code file via the web interface.
3

Chunking: The backend's Smart Chunker parses the file into a list of complete, logical functions using AST.
4

Initiate Streaming: The server opens a persistent connection back to the browser to begin streaming real-time updates.
5

Parallel Analysis: The ThreadPoolExecutor assigns each function chunk to a worker thread, which sends an analysis request to the AI API concurrently.
6

Partial Results Stream: As each worker thread gets a response, the result is immediately streamed to the browser and displayed.
7

Gather Summaries: Once all threads complete, the backend collects the lightweight, structured summaries for each function.
8

Construct Final Prompt: A single, comprehensive prompt is built using the collected summaries to ask for a high-level verdict.
9

Final API Call: One last request is sent to the AI API to get the overall functionality, artifacts summary, and final verdict.
10

Final Result Stream: The final report is streamed to the browser, populating the summary cards, and the connection is closed.

AI-Powered Executable File Analyzer

Detailed System Architecture

User's Browser (Client)

Cloud Run Service

Generative AI API