In today's increasingly remote working style, organization’s messaging system, whether it's email or chat, contains lots of invaluable institutional knowledge. However, because these data are often unstructured and scattered, they are usually buried in the organization’s data ecosystem and are hard to search and extract value. Fetcher is a chatbot that integrates into popular chat platforms such as Discord and Slack to seamlessly help users find relevant people and documents to save them from endless frustrating search. It does this by semantically searching chat messages to find the most relevant results and help to deliver actions that leads to a peace of mind. Fetcher differs from traditional keyword search engines in that it searches by the meaning of the query, not just by keywords. It also enables multi lingual search, so that global teams can more quickly find important information even when language is a barrier. Since Fetcher searches in the embedding space, this search engine can extend to multi modal modes that includes audio and images. Fetcher works by collecting a chat channel’s history and embedding them using Cohere’s Embed API, then saving the embeddings to Qdrant’s vector search engine. When a new query comes in, Fetcher embeds the query and searches against the vector database to find the most relevant results, which can then feed into Cohere’s Generate API to summarize the message thread to kick start new conversations. Fetcher offers 3 commands, /fetch, using vector similarities search to find relevant chat messages. /discuss, summarize a message thread, and kick start a conversation with a channel number. /revise, a sentence correction tool similar to Grammarly, allows user to send professional sounding messages.
17 Mar 2023
.png&w=828&q=75)
Podcasts are an excellent source of knowledge. But they can be too long and hard to pay attention to it the entire time. What if there is a more intuitive way to search for podcasts and also for info within podcasts? This is where our product comes into play. Key highlights 1. Searching for podcasts suited to your taste 2. Searching for answers within a podcast itself by asking it queries and without listening 3. Marking exactly where the answer is and summarising it. 4. Telling user what queries this podcast answers Major Uplifts: 1. Generating queries for dialogues in transcript using the prompt - "Generate 5 questions for the following passage {passage}" 2. Training a classifier using cohere api using the generated queries and dialogues 3. Highly scalable architecture 4. Podcast is just an example. Most documentation (python libraries, eth doc) have only keyword search. It is possible to scrape the documentation and build an index for a search engine using our architecture easily.
24 Dec 2022
Distiller condenses information shared during meetings into bit-sized summaries and provides inspirations and actionable plans to drive projects forward productively. It transcribes long discussions into searchable transcript, summarizes content into easily consumable forms, provide action items and follow-up questions to push the project forward, and generate metaphors and images to promote more brainstorming.
17 Dec 2022
Smarty quiets the noise around you so you can work peacefully to get things done. It is a personalized task planner that evaluates the importance and urgency of your tasks, then prioritizes and batches them rigorously so you can focus on accomplishing what matters the most.
5 Dec 2022
.png&w=256&q=75)
.png&w=640&q=75)
.png&w=640&q=75)
