CASE STUDY / KINDO
Secure and compliant AI management
Kindo is a forward-thinking startup operating in the AI sector, specializing in Large Language Models (LLMs) like ChatGPT and cloud technologies.
01 / THE CLIENT
Who they are.
Dedicated to making AI accessible to businesses, the company offers a SaaS solution that integrates with various platforms, including Dropbox, Google Drive, Jira, Linear, accounting software, CRM tools, and more, all supported by merge.dev.
The company sought our expertise in staff augmentation to enhance their front-end and back-end development capabilities.
02 / THE CHALLENGE
What they needed.
Initially, the client’s need for a front-end developer focused on React Native mobile development marked the commencement of our collaboration.
Beyond this, the challenges expanded into the intricate realm of backend development, necessitating the incorporation of advanced Large Language Models (LLMs) like gpt4, gpt3, claude 100k, claude v2, claude Instant, falcon, and palm. The integration of these LLMs was pivotal for the client’s objective of making AI accessible to businesses through their SaaS offering.
To enhance the AI capabilities, our team delved into improving data organization and embeddings, adopting technologies such as Pinecone Store for efficient storage and retrieval of embeddings. The introduction of a synchronized background worker became imperative to pick up real-time changes in documents from platforms like Dropbox, Google Drive, and various ticketing systems. This worker tokenized and embedded each new or modified item, creating a dynamic index for the LLMs to gain context as per user requests.
The challenges were multifaceted, ranging from enhancing AI capabilities through LLM integrations to optimizing data organization, embedding mechanisms, and token balance management. The technical intricacies involved advanced technologies and a meticulous approach to ensure the integration of diverse systems and the successful delivery of an advanced AI solution.
Some of our services included:
- Front-end development using Next.js, React and React Native
- Back-end development using Node.js
- Integration of various LLMs, including gpt4, gpt3, claude 100k, claude v2, claude Instant, falcon, palm
- Implementation of merge.dev unified models for ticketing, storage, accounting, human resources, marketing, and CRM.
03 / RESULTS
Where it ended up.
We created alarms and metrics using DataDog to visualize events and metrics effectively. All environment variable management was streamlined into Doppler for enhanced configuration management. An embedding mechanism utilizing Pinecone Store was introduced, and a synchronized background worker was developed to tokenize and embed changes in documents, tickets, and CRM entries, making them readily available for the LLM.
GET IN TOUCH
Tell us where you are stuck.
If any of this looks like the problem you have, a thirty-minute call is usually enough to know whether we are a fit.
Book a 30-minute call