Inspiration
Sales Representatives spend many hours trying to log data to their CRM fields to ensure that no deal data is left behind. This is a monotonous task with no value-prop for the reps as the they could utilize their time selling to more organizations. This inspired me to create this product.
What it does
The product captures data from a meeting and uses either OpenAI models to create embeddings. These embeddings are then stored in a PineCone index and retrieved when a rep wants to update their CRM. The aim is to create a retrieval-augmented generative AI model that automatically fills in the CRM for the rep, saving tons of time and increasing productivity.
How we built it
The rep inputs a meeting. This meeting is transcribed and the data is embedded using OpenAI's text-embedding-ada-002 model. These embeddings are then stored in a PineCone index. PineCone acts as long-term memory for OpenAI's models. The rep defines and inputs the fields that are to be filled. These fields are then used to query relevant vectors from the PineCone index and then passed onto OpenAI's GPT-4 model to produce the final answer.
Challenges we ran into
The biggest challenge we faced was retrieving relevant portions of the meetings that capture enough context so that the OpenAI GPT-4 model can comprehend the context and produce coherent answers. To tackle this problem, we tried different types of filtering mechanisms while querying the index which are able to capture the relevant portions of the meetings.
Accomplishments that we're proud of
I'm proud of the fact that the architecture is able to generate coherent answers for most of the fields.
What we learned
I learnt about different types of techniques in Information Retrieval and how these techniques should be used based on the different types and modalities available.
What's next for Automated CRM Assistant
For now, we only have support for MEDDPICC fields. Next, we also want to support other frameworks like SPICED etc. We also want to add the feature for extending these fields to custom fields that maybe used an Enterprise to capture metrics related to their deals.
Log in or sign up for Devpost to join the conversation.