Pygma

Generates original social media content using LLM inference (likely GPT-based) with automatic adaptation to platform constraints (character limits, hashtag conventions, media requirements). The system accepts user briefs, brand context, or content topics and outputs formatted posts ready for immediate scheduling. Architecture likely involves prompt engineering templates that inject platform-specific rules and brand voice parameters into the generation pipeline.

Manages post scheduling across multiple social platforms (Twitter, Instagram, LinkedIn, TikTok, Facebook) with a unified calendar interface. Posts are queued with scheduled publish times and automatically distributed to each platform's native API at the specified moment. The system handles platform-specific authentication (OAuth tokens), rate limiting per platform, and retry logic for failed publishes. Architecture uses a task queue (likely Celery or similar) to trigger publishes at exact timestamps.

Allows users to define brand voice parameters (tone, vocabulary, style, values) that are injected into the LLM prompt during content generation. Users provide examples of on-brand content, tone descriptors (professional, casual, humorous, etc.), and brand values, which are encoded as system prompts or few-shot examples. The generation pipeline uses these parameters to constrain output style, though effectiveness depends on prompt engineering quality rather than model fine-tuning.

Provides a unified calendar interface showing all scheduled posts across platforms with drag-and-drop rescheduling, bulk editing, and content preview. The calendar supports month/week/day views and displays posts color-coded by platform. Users can batch-select posts, apply changes (reschedule, edit, delete), and preview how content will appear on each platform before publishing. Architecture uses client-side state management (React/Vue) with backend sync for persistence.

Tracks engagement metrics (likes, comments, shares, impressions, reach) for published posts by querying platform APIs (Twitter Analytics API, Instagram Insights API, LinkedIn Analytics API). Metrics are aggregated in a dashboard showing post-level performance, engagement trends over time, and basic comparisons (best-performing post type, optimal posting time). Architecture uses scheduled API polling (daily or weekly) to fetch metrics and store in a time-series database for historical analysis.

Implements a freemium model with restricted posting limits (e.g., 5-10 posts/month free, unlimited on paid tier) enforced via quota tracking in the backend. The system counts published posts against the user's monthly allowance and blocks publishing when quota is exhausted, with upgrade prompts to paid plans. Quota resets on a monthly billing cycle. Architecture uses a simple counter in the user database with monthly reset logic.

Handles OAuth 2.0 authentication flows for connecting social media accounts (Twitter, Instagram, LinkedIn, Facebook, TikTok) to Pygma. The system stores encrypted OAuth tokens, manages token refresh (some platforms require periodic refresh), and handles authentication errors gracefully. Architecture uses a secure token vault (likely AWS Secrets Manager or similar) with automatic refresh logic triggered before token expiration.

Generates content topic ideas and post concepts based on user input (industry, audience, brand), trending topics, or historical post performance. The system uses LLM inference to brainstorm content angles, hooks, and themes that align with the user's brand and audience. Ideas are presented as prompts that can be directly fed into the post generation capability. Architecture likely uses prompt templates that inject industry context and trending data into the LLM.