1.4 million
of call data processed
Hierarchical LLM for ultra-high data volume processing.
of call data processed
increase in marketing team productivity
Product discovery
Chatbots and text-based human interfaces
Knowledge base search, RAG, and vector databases
Data extraction from unstructured data sources
Workflow supercharging (e.g., customer service inquiry)
Development
A multi-billion US-based business outsourcing enterprise offering voice and non-voice services for a range of customer touchpoints - customer care, retention, onboarding, and collections. The project involved implementing an Interactive Reporting Experience on millions of call transcripts collected across clients’ call centers all around the world.
To evaluate key performance indicators (KPIs), such as handle times, customer satisfaction, and compliance, it was necessary to automate tasks associated with analyzing and evaluating call transcript data.
We started with a 3-week discovery to specify the areas for improvement and analyze the business impact of automation and the use of AI on client’s processes. We conducted intensive discovery sessions that included user interviews, feasibility discussions, and infrastructure reviews to understand the client’s starting point and ability to implement AI to analyze and evaluate large quantities of data.
After summarizing all our findings, we identified multiple opportunities for improvement. Subsequently, we assessed various state-of-the-art large language models (LLMs) and provided recommendations on which one to use given various constraints (including privacy, accuracy, speed, budget, etc.).
Finally, we developed a proprietary hierarchical model structure to bypass several known limitations of leading LLMs and allow for ultra-large context processing.
Through our deep AI expertise, we overcame several known limitations of state-of-the-art language models. We built a custom solution that allowed the client to process and analyze >100x more significant amounts of call data than before. With this solution, the client achieved significant value uplift from efficiency gains (i.e., reduction in analyst time) and topline uplift (e.g., mitigation of root causes for customer dissatisfaction, identification of upsell opportunities, etc.).
Next.js (React.js)
Typescript
Apollo GraphQL Client
Python
Langchain for agents
FastAPI
Celery for background task execution
Render
44 Montgomery St, San Francisco, CA 94104
