Documenting code, one commit at a time.
In the devlog-ist/landing project, we're focused on delivering high-quality content. A crucial part of this is ensuring that AI-generated content meets our standards before it's published.
AI models, particularly when generating longer pieces of content, can sometimes be cut short due to token limits or other constraints.
This post explores the implementation of cost tracking and budget limits for AI operations within the devlog-ist/landing project. We'll examine how per-request costs are tracked in EUR, how pricing is stored per model, and how monthly cost limits are enforced per plan.
The system now tracks the cost of each AI operation in EUR. This involves:
This post details a small but important change in how we generate content for sharing on LinkedIn, focusing on aligning the tone with the platform's user experience.
The primary goal was to shift the voice of automatically generated LinkedIn share posts to a first-person perspective. Since the posts are published from the user's profile, using "I" instead of third-person references
This post details the recent enhancements to the content generation process within the devlog-ist/landing project, focusing on the integration and management of LinkedIn prompts for improved content quality and platform-specific tailoring.
The primary objective was to enhance the content generation workflow by incorporating LinkedIn-specific prompts, allowing for more targeted and
We recently added a "Share on LinkedIn" feature to our landing page project, devlog-ist/landing. The goal was to increase user engagement and expand the reach of user portfolios by enabling seamless sharing on LinkedIn. This post details the implementation process, covering AI-powered post generation, direct publishing via the LinkedIn API, and considerations for different user
When generating content for LinkedIn using AI, it's crucial to tailor the prompts for conciseness and engagement. The goal is to create posts that fully encapsulate the idea within LinkedIn's character limit, avoiding truncation and maximizing impact.
Key optimizations include instructing the AI to generate short, focused content (around 2500 characters), structured in 3-5 paragraphs.
Working on the landing project, which is focused on creating a compelling user experience, we've recently enhanced our testing strategy to ensure the reliability of our GitHub API integrations.
The goal was to catch potential issues arising from changes to the GitHub API, preventing silent failures in our application. We've implemented a suite of integration tests that validate the structure of
Improving the auditability of AI interactions is crucial for maintaining security and control. A recent update focuses on preventing the exposure of raw code to AI models, enhancing data security, and providing better insights into flagged code changes.
Previously, raw git diffs were sent to AI models for analysis. This approach, while providing detailed context,
This post discusses recent improvements to our application, focusing on enhanced security measures and smarter code generation capabilities.
We've introduced a 'safe mode' feature, giving tenants more control over security audits during post generation. By default, safe mode is enabled, ensuring all generated content undergoes a thorough security check.
Improving the way we audit code changes is crucial for maintaining security and stability in our applications. Recently, we transitioned from feeding raw Git diffs directly to our AI analysis tools to using structured summaries. This shift significantly enhances auditability and reduces the risk of exposing sensitive information.
Sending raw diffs to AI models
