Effective Date: 6 May 2026
Hatch uses artificial intelligence to deliver several features of the Service. This notice explains, in plain language, where AI is used, what it does, what its limits are, and what choices you have. We publish this notice to comply with transparency obligations under emerging AI regulation (including the EU AI Act, Article 50, in force from 2 August 2026), data-protection laws, and our own values.
Where AI is used in the Service
| Feature | What it does | Models / providers used |
|---|---|---|
| Document understanding | Extracts text from your uploaded documents and breaks it into smaller chunks for indexing | OpenAI text embeddings (currently text-embedding-3-small, 1536 dimensions) |
| Learning material generation | Generates structured chapter plans, articles, summaries, and key takeaways from your uploaded documents | OpenAI chat models (currently gpt-4o) |
| Quiz generation | Generates multiple-choice quizzes that test the content of your learning materials | OpenAI chat models (currently gpt-4o) |
| Chatbot | Answers questions from workers, drawing on relevant chunks of your uploaded documents (retrieval-augmented generation, "RAG") | OpenAI chat models (currently gpt-4o); pgvector for retrieval |
| Translation | Translates generated content into supported languages, including Georgian | OpenAI chat models |
| Podcast generation | Converts a generated podcast script into spoken audio | ElevenLabs text-to-speech (eleven_monolingual_v1 for English, eleven_v3 for Georgian) |
What AI does NOT do
- It does not make employment decisions about you. Quiz analytics and learning progress are intended for measuring learning outcomes. Our contract with your employer prohibits using these as the sole or primary basis for hiring, promotion, termination, compensation, or disciplinary decisions.
- It does not access data outside your organization. The chatbot answers from documents your own organization has uploaded; it does not search the public internet, your email, or any other system.
- It does not learn from your conversations. We do not use your data, your chats, or your employer's documents to train AI models. Where our AI providers offer training-data opt-outs, we use them.
- It does not decide who sees what content. Access controls (which worker sees which materials) are configured by your HR administrator, not by AI.
Known limits
You should be aware of the following:
- AI can be wrong. Generative AI models can produce confident-sounding answers that are factually inaccurate, incomplete, or out of date. This phenomenon is sometimes called "hallucination". You should always verify important information with a knowledgeable human (your manager, your HR team, or an authoritative source) before relying on it for decisions with legal, financial, medical, safety, or similarly significant consequences.
- AI inherits biases from its training data. Generated content may reflect cultural, linguistic, or other biases present in the model's training data. We work to reduce these but cannot eliminate them.
- Translation quality varies. Automated translation, especially into less-resourced languages, may miss nuance or use awkward phrasing. Critical communications should be reviewed by a fluent human.
- Synthetic voice is not the speaker's real voice. Podcast voices ("Alex", "Jordan", and the Georgian-language voices) are synthetic. They do not represent any specific real person and are not recordings of any human reading the script.
- The chatbot's knowledge is bounded by your uploaded documents. If something isn't in your organization's uploads, the chatbot can't answer accurately. It may try anyway. Treat unsourced answers with skepticism.
What you should do
- Treat AI outputs as a draft or starting point, not as a final source of truth.
- Read the citations or document references shown alongside chatbot answers, where available.
- Tell your HR administrator or manager if you spot a clearly incorrect or harmful AI output. We log and review these reports.
- If you are an HR administrator: review the content generated for your workforce.
How AI generation works (high level)
- Your HR administrator uploads documents (e.g., handbooks, training materials, policies).
- We extract text and split it into chunks.
- Each chunk is converted into an embedding (a list of numbers representing its meaning) and stored in a vector index.
- A planning step uses an AI model to organize the documents into chapters and topics.
- A generation step produces an article, summary, quiz, and (optionally) podcast script for each chapter.
- When a worker asks the chatbot a question, the system finds the most relevant chunks from your organization's documents and passes them, along with the question, to an AI model to generate the answer.
- Generated audio is created by sending the script to a text-to-speech provider.
The AI providers we use are listed on our Sub-processors page. Personal data flows are described in our Privacy Policy.
Your rights
Under applicable data-protection law, you may have rights including the right not to be subject to a decision based solely on automated processing that produces legal or similarly significant effects on you. As described above, we do not make such decisions through this Service. If you have questions or want to exercise any rights, contact your employer's data protection contact, or write to us at hatch.officiall@gmail.com.
Changes
We may update the AI models or providers we use to improve quality, reduce cost, or address security or compliance considerations. Material changes — including changes to providers — will be reflected in this notice and on our Sub-processors page.
Last reviewed: 6 May 2026 Contact: hatch.officiall@gmail.com