CAEngine (insights_engine)

Analyzes a GitHub repository, finds the code that shows real engineering skill (transactions, idempotency, event-driven messaging, auth guards, caching) and turns it into ready-to-publish content: a LinkedIn post, an X thread, a code screenshot and an architecture diagram.

Status: personal project, work in progress. Content generation works end to end. Publishing modules for X, Bluesky, dev.to and LinkedIn are implemented but disabled by default (the calls are commented out in main.py).

How it works

1. Scan the target repository and rank its files by relevance (repo_scanner.py).
2. Audit seniority: assigns a rank and a score based on the patterns found.
3. Detect concepts from five categories: concurrency, resilience, event-driven patterns, security and performance.
4. Extract the best snippet with local heuristics, falling back to an LLM when the result is weak.
5. Generate content with LLMs (Gemini and Groq): a LinkedIn post and an X thread.
6. Render visuals: a code image (Ray.so through Playwright) and an architecture diagram (Mermaid).
7. Save a media kit per topic under content_factory/.

Output example:

content_factory/
└── Kit_1_serie_1_real-world_concurrency_2026-06-09/
    ├── linkedin.md
    ├── x_thread.md
    ├── 1_authority_shot.png
    └── 2_architecture.png

Tech stack

• Python
• Google Gemini (google-genai) and Groq for content generation
• Playwright for rendering code screenshots
• Platform APIs: X, Bluesky, dev.to, LinkedIn (disabled by default)

Getting started

git clone https://github.com/VittoLym/insights_engine.git
cd insights_engine
python -m venv .venv
.venv\Scripts\activate          # Windows (use source .venv/bin/activate on Linux/macOS)
pip install -r requirements.txt
playwright install chromium
cp .env.example .env            # then fill in your keys

Set the repository to analyze in main.py (the SOURCE variable at the bottom of the file), then run:

python main.py

The target repository is currently hardcoded. Accepting it as a command-line argument is on the roadmap.

Configuration

Copy .env.example to .env. At minimum you need GEMINI_API_KEY and GROQ_API_KEY to generate content. The X, Bluesky, LinkedIn and dev.to variables are only needed if you enable publishing.

Enabling publishing

In the __main__ block of main.py, the publish calls are commented out. Uncomment only the platforms you have configured:

# publish_linkedin(linked_post, [pngPath])
# publish_thread_bluesky(x_thread)
# publish_thread_x(x_thread)
# publish_devto(...)

Review the generated content in content_factory/ before you publish anything.

Privacy note

To render visuals, code snippets are sent to third-party services: Ray.so (code image) and mermaid.ink (diagram). Do not run the tool on private or confidential repositories unless you are comfortable with that.

Roadmap

• Accept the target repository as a CLI argument
• Split main.py into modules (scanner, content, media, publishers)
• Remove unused code left over from earlier iterations
• Unit tests for scoring and snippet extraction
• Meta (Instagram / Facebook) support
• Scheduled publishing

How I used AI in this project

## How I used AI in this project
- **What I did myself:** the research, the design of the tool, the setup of every API account and key (Gemini, Groq, X, Bluesky, dev.to, LinkedIn) and the validation of the code.
- **What I delegated:** AI assistants wrote the repetitive parts of the code.
- **How I reviewed it:** I read and validated the generated code and ran the tool end to end, checking the content it produces before publishing anything.

Author

Alexander Assón, Backend / Full Stack Engineer, Mendoza, Argentina. LinkedIn · GitHub