OrbitNote AI

Inspiration

What it does

In small companies, the data analytics department can be quite chaotic. You may have an analyst whose job is to crunch whatever numbers they can find. They will produce reports which then inform the business executives about the state of their business. However, building an organized and reliable data store is often too difficult of a task for a small company to take on, so the analyst either becomes a bottleneck, or data is not consulted as often as it should.

OrbitNote aims to be a repository of knowledge for such a company. The platform allows users to upload Jupyter Notebooks and share them with the business team. OrbitNote AI is an addon to OrbitNote which enables the platform to run RAG-based queries on the data users uploaded.

How we built it

The POC is using a few sample Juptyer Notebook reports. There's a collection of scripts that will retrieve information and data from those reports using fine-tuned prompts to ChatGPT. Files with those facts and data are then fed to Vectorize. End user has a CLI script that they can use to ask questions to their knowledge repository.

Challenges we ran into

AI not understanding the concepts of temporality of data very well (vectorize not able to accurately pull data with "first of February" or "end of March" data)

Accomplishments that we're proud of

POC is working!

Example output

orbitnote-cli -q "What was our Daily Active User in mid-Feb"            

Vectorize query: 
What was our Daily Active User count during the period from February 10 to February 20, 2025?



Answer: 
The Daily Active Users (DAU) for Company X's web application on February 15, 2025, was 180 [source=11bd1003-7ef6-4ca1-ba7b-f041e56cfb4f&link=sample-1.txt](11bd1003-7ef6-4ca1-ba7b-f041e56cfb4f/f2d7b6aa-2e26-4cd6-8afa-2cdacee1662a/sample-1.txt).



Conclusion: 
What was the Daily Active User count for the entire mid-February period in 2025?



Revised: 
The Daily Active Users (DAU) for mid-February 2025 were as follows: on February 14, 2025, there were 230 DAUs, and on February 15, 2025, there were 180 DAUs [source=11bd1003-7ef6-4ca1-ba7b-f041e56cfb4f/f2d7b6aa-2e26-4cd6-8afa-2cdacee1662a/sample-1.txt](11bd1003-7ef6-4ca1-ba7b-f041e56cfb4f/f2d7b6aa-2e26-4cd6-8afa-2cdacee1662a/sample-1.txt).

or a more complex query - specifically this one required the agent to use multiple data sources because each report only holds data for 30 days.

orbitnote-cli -q "Compare DAU for first week of Feburary and last week of March"

Poe => python orbitnote_ai/query_data.py -q 'Compare DAU for first week of Feburary and last week of March'

Vectorize query: 
What was the daily active users (DAU) for the first week of February 2025 and the last week of March 2025?



Answer: 
To compare the DAU (Daily Active Users) for the first week of February and the last week of March 2025, we can refer to the data provided:

1. **First Week of February 2025:**
   - February 1: 360
   - February 2: 390
   - February 3: 420
   - February 4: 450
   - February 5: 490
   - February 6: 530
   - February 7: 510
   - The average DAU for this period is approximately 450 ([source](11bd1003-7ef6-4ca1-ba7b-f041e56cfb4f/f2d7b6aa-2e26-4cd6-8afa-2cdacee1662a/sample-1.txt)).

2. **Last Week of March 2025:**
   - March 25: 230
   - March 26: 220
   - March 27: 250
   - March 28: 260
   - March 29: 250
   - March 30: 240
   - March 31: 230
   - The average DAU for this period is approximately 240 ([source](11bd1003-7ef6-4ca1-ba7b-f041e56cfb4f/f2d7b6aa-2e26-4cd6-8afa-2cdacee1662a/sample-5.txt)).

From this comparison, DAU was significantly higher during the first week of February compared to the last week of March.



Conclusion: 
No further action required.

What we learned

First brush with Vectorize and building more complex Agent-like networks (format user question, delegate question to Vectorize, generate answer, review answer for completion, consider asking a follow-up question to Vectorize, process follow-up query)