Knowledge workers spend a majority of their time on repetitive menial tasks which prevent them from focusing on thoughtful work which they thrive to do. After talking to several knowledge workers in consulting, banking, marketing, etc we realized that a majority of these undesirable tasks involve locating sources of information and pulling data from them. And thus we built manna to solve this problem.

Manna is a browser extension that let’s users quickly retrieve information from their documents. When working on a document or email users can press cmd+e to call manna which analyses the current context and recommends them the most relevant pieces of information from their online and offline files. They can also query manna to find more information.

How we built it

As a proof of concept, we scraped and indexed public SEC filings. Then we developed search algorithms that would direct the user the correct part of the document when queried.

Coming up with a good search algorithm was hard, we tried Word2vec, Whoosh and bi-direction transformers (BERT) and in the end we did multiple levels of search granularity, starting with a high level search to find related chunks of documents and then a finer grain search for finding desired phrases.

Based on the page the user is at we create a context vector and combine it with possible search queries that a user might have. Then we feed this request to an initial coarse search that uses Whoosh to narrow down which section of which document to look into. Then we used Word2Vec and bi-directional transformer models (BERT) to find the sentences that are related to what the user wants to see.

On the front-end side we developed a chrome extension that has a search UI that queries a flask server and display the most relevant search results.

In the end, we successfully developed a ready to use chrome extension that let’s us search our documents much faster than

What’s next

Despite the universality of our problem space, we think private equity firms are the perfect beach head to start with. A typical private equity analyst receives 1000 documents every few weeks on Intralinks or box and has to make sense of the data and analyze it in the next two weeks. The current products have a frustrating user experience mainly because of an incentive misalignment in the market. The customers of cloud drive services are banks that are trying to sell their companies not not the buyer private equity firms that need to analyze the documents. Because of this the current solutions serve the needs of the seller and not the buyer and thus don't have any notion of fast context-based search. Therefore, we think there is a market gap and an easy to adopt local chrome extension can help them reduce their current 8-10h time of understanding the structure of documents in a data-room to less than 4 hours.

We believe this can be a first step towards a self-driving browser that automates thoughtless parts of work and lets us focus on what we are best at.

Check out our work!

GitHub - behzadhaghgoo/manna-yc

Share this project: