Essai: A Stateless, Privacy-First Academic Writing Assistant

Essai is an academic writing assistance system released by JohnKing376 on GitHub on June 5, 2026. Its core design principles are stateless architecture and privacy-first. The system submits user text to large language models (LLMs) via dynamic prompt engineering and returns structured academic writing evaluation results, aiming to address the data storage privacy risks of traditional AI writing tools.

Essai adopts a stateless architecture, which differs from the session state maintenance mode of traditional web applications. Each request contains all processing information, and the server does not save client state. This design brings multiple advantages: 1. Strong horizontal scalability, suitable for computationally intensive academic evaluation tasks; 2. Simplified failure recovery—instance failures do not affect independent requests; 3. Flexible deployment (containerized/Serverless), and lays the foundation for privacy protection (user data is only temporarily stored in memory).

Essai takes privacy protection as its core goal, with strategies including: 1. Minimal data collection: only processes text actively submitted by users, no extra metadata is collected; 2. No persistent storage: user content exists only during request processing and is immediately released after completion; 3. Proxy mode: acts as a proxy between users and LLMs, enabling data desensitization/anonymization; 4. Transparency: clearly informs users that data will be sent to third-party LLM service providers.

Essai implements core functions via dynamic prompt engineering: 1. Input analysis: identifies text type (plain text/specific format) and structural features; 2. Context construction: dynamically generates prompts containing role settings, evaluation criteria, and output formats based on input type and evaluation needs (e.g., academic norm checks, argument logic analysis, etc.); 3. Structured output: ensures LLMs return structured results such as scores and improvement suggestions; 4. Multi-model support: can connect to different LLM service providers to adapt to different task requirements.

Essai designs evaluation dimensions for academic scenarios, covering: 1. Structural integrity: checks whether paper sections are complete and logical connections are smooth; 2. Argument quality: evaluates the clarity of arguments, sufficiency of evidence, and rigor of reasoning; 3. Academic norms: verifies citation formats (APA/MLA, etc.), completeness of references, and originality; 4. Language expression: checks the appropriateness of academic language and grammatical errors; 5. Technical accuracy: verifies the correct use of professional terms and concepts.

Essai's target users and scenarios include: 1. Academic writers (graduate students/researchers): writing quality checks and improvement suggestions; 2. Educational institutions: teachers evaluate student papers in batches to improve correction efficiency; 3. Journal editors: quickly screen submission quality in the initial review stage; 4. Language learners: non-native writers improve their academic English skills.

Privacy and Security Boundaries: Essai does not store data, but users need to trust the data policies of third-party LLM service providers, and should avoid submitting highly sensitive content; data transmission must be encrypted (HTTPS/TLS). Comparison with Similar Tools: Compared to traditional tools, Essai is stateless and does not store data, ensures privacy at the architecture level, may support anonymous use, focuses on academic evaluation, and its cost may be pay-as-you-go.

Essai achieves a balance between privacy protection and functional practicality. The stateless architecture is a technical manifestation of its privacy commitment, and dynamic prompt engineering effectively leverages LLM capabilities. Recommendations: 1. Pay attention to the data processing policies of LLM service providers; 2. Avoid submitting unpublished research results or sensitive information; 3. Use its structured evaluation to improve academic writing quality. This design concept may provide a reference for privacy protection in AI tools.