Developers are increasingly frustrated with the homogenization of AI-generated text, where unique author voices are flattened into predictable corporate drones. Enter InkFlow, a new open-source project submitted to the Hacktoberfest Open-Source AI Challenge that aims to solve this by building an adaptive writing twin. The tool focuses on preserving individual style while actively detecting and removing common AI clichés, offering a practical alternative to generic LLM polishing tools.

The Style Genome Architecture

At the core of InkFlow is the 'Personal Style Genome,' a feature that requires users to complete a two-minute linguistic calibration exercise. This process builds an eight-dimensional profile capturing specific metrics like formality, syntactic density, lexical entropy, and punctuation signatures. Unlike simple prompt engineering that asks for a 'tone,' InkFlow tracks contraction ratios and clause chaining to create a structured baseline of how a specific user writes, ensuring that subsequent edits align with their actual habits rather than a generic approximation.

Dual-Model Inference for Reliability

From an infrastructure perspective, the project demonstrates a robust approach to handling LLM latency and availability. The backend, built on FastAPI and Python 3.11+, utilizes Groq for inference but implements a dual-tier strategy. It primarily calls openai/gpt-oss-120b for complex stylistic transformations but automatically falls back to openai/gpt-oss-20b if the primary model encounters timeouts or rate limits. This resilience is critical for a side-hustle product that cannot afford downtime due to single-provider failures, and it leverages open-weight models to keep costs manageable for the freemium business model.

Sequential Pipeline and Semantic Safety

The rewriting engine operates through a strict four-stage pipeline designed to prevent 'semantic drift,' a common failure mode in AI editing where factual accuracy is lost during stylistic changes. Stage one generates a base draft, stage two applies the Style Genome and removes slop, stage three verifies semantic consistency, and stage four performs a quirk-preserving grammar pass. The system specifically protects intentional punctuation choices, such as em-dashes or semicolons, which standard grammar checkers often strip away, thereby maintaining the author's unique rhythm while correcting errors.

Key Takeaways

  • InkFlow uses an eight-dimensional 'Style Genome' to capture specific linguistic traits like burstiness variance and tone affinity, moving beyond simple tone prompts.
  • The infrastructure employs a fallback mechanism using openai/gpt-oss-20b when openai/gpt-oss-120b fails, enhancing reliability for users on the free tier.
  • The project is monetized via a ₹39/month Pro tier using Razorpay, targeting students and creators who need personalized writing assistance without high API costs.
  • A dedicated semantic fidelity check ensures that aggressive cliché removal does not accidentally delete factual claims or numerical data from the source text.

The Bottom Line

InkFlow proves that the next wave of dev tools won't just generate text, but will actively defend authorial voice against the entropy of large language models. By combining open-weight inference with a structured style profile, it offers a viable path for indie hackers to build sustainable, privacy-respecting writing aids.