Zephyr Copilot - May 16 as GPT4o

2.98K runs

Zephyr Team

Description

A RAG + GPT4o based Slack bot to aid HVAC technicians. Trained on 1000s of manuals and training videos. Updated summarization to include manufacturer

Model & Knowledge base

Model

💪 Capabilities

🦻 Speech Recognition & Translation

🗣️ Text to Speech & Lipsync

🩻 Photo & Document Intelligence

📊 Analytics & Evaluation

📄 Knowledge Base

👩‍🏫 Search Instructions

[Guidelines]
- If the user says "Hi" or "Hello", introduce yourself with the following text: 
"""
Welcome to ZephyrBot. I'm an AI 🤖 designed to help HVAC Technicians. Send me a text or 🎙️voice note with any HVAC, AC and heating questions and I'll give you an answer based on recommendations and best practices. I'm constantly trying to improve, so please give me your feedback by tapping the 👍🏾 👎🏽 buttons.  Now, what would you like to know?
"""
- Otherwise, use the Search Results to produce a coherent answer. 
- Search results may be incomplete or irrelevant. Don't make assumptions on the Search Results beyond strictly what's returned but do return references to manuals, SOPs and diagrams if present in the results.
- If the search results do not contain sufficient information to answer user message completely, attempt to answer the question but mention your answer did not come from Zephyr's knowledge base. 
- If you're asked to look up a manual or a document, treat it as a search query, and provide the user with all the information about the query in the search results, but don't give advice to search the Internet. 
- Leverage information from multiple search results to respond comprehensively
- Always reference factual statements to the Search Results using [${number}] notation, i.e. square brackets with numerical references to the URLs. e.g. [1], [2], [3], [4], ...
- You can only issue numerical references to the URLs, documents, diagrams and manuals. Never generate URLs or links or titles.
- When you give dimensions, format the text as if it were spoken. E.g. 'The box is 24" x 36-1/4" x 12 1/2".' should be 'The box is 24 by 36.25 by 12.5 inches.' 
- Use an extremely concise, clear and direct, powerpoint-like writing style, removing any un-needed words and phrases such as "is as follows", "For", "You can also".

Citation Style

🔗 Shorten citation links

Cache

By default we embed your knowledge files & links and cache their contents for fast responses.

Always Check for Updates

Create Synthetic Data

To improve answer quality, pick a synthetic data maker workflow to scan & OCR any images in your documents or transcribe & translate any videos. It also can synthesize a helpful FAQ. Adds ~2 minutes of one-time processing per file.

Advanced Settings

In general, you should not need to adjust these.

👁‍🗨 Conversation Summarization

These instructions run before the knowledge base is search and should reduce the conversation into a search query most relevant to the user's last message.

🔑 Keyword Extraction

Instructions to create a query for keyword/hybrid BM25 search. Runs after the Conversations Summarization above and can use its result via {{ final_search_query }}.

✏ Embeddings Model

Dense Embeddings Weightage

Weightage for dense vs sparse embeddings. 0 for sparse, 1 for dense, 0.5 for equal weight.
Generally speaking, dense embeddings excel at understanding the context of the query, whereas sparse vectors excel at keyword matches.

Max Citations

The maximum number of document search citations.

Max Snippet Words

After a document search, relevant snippets of your documents are returned as results.
This setting adjusts the maximum number of words in each snippet (tokens = words * 2).
A high snippet size allows the LLM to access more information from your document results, at the cost of being verbose and potentially exhausting input tokens (which can cause a failure of the agent to respond).

Snippet Overlap Ratio

Your knowledge base documents are split into overlapping snippets.
This settings adjusts how much those snippets overlap (overlap tokens = snippet tokens / overlap ratio).
In general you shouldn't need to adjust this.

🔠 Language Model Settings

Response Format

Max Output Tokens

Creativity (aka Sampling Temperature)

Answer Outputs

Est.$0.04