Sep 2023 – Present · 2 yrs 6 mos · Bengaluru, Karnataka, India · Hybrid

• TL;DR: Primarily work on P0 features and bugs. Continuously striving towards and creating differentiated value for the company.
• AI voice calling engine
• Built the majority of the speech service which handles real-time conversations with a borrower with a median response time of < 1 second.
• Built the entire distributed calling intelligence system, managing data storage, call scheduling, and borrower profiling, with a capacity to process 4 million calls per day.
• Delegated these parts of the system to other engineers in the team after stabilising them.
• Contributing in their continuous development and maintenance through code reviews, architectural decisions and working on P0 issues and features.
• LLM Fine-tuning
• Fine-tuned open-source LLMs on proprietary data for debt-collection conversations to increase quality and control over the model output while also becoming compliant w.r.t. data residency.
• Fine-tuned private LLMs to reduce costs for multiple LLM-dependent workflows by 16x.
• Oversaw the fine-tuning of open-source LLMs on 7 additional workflows by guiding engineers in the team. Helping in data synthesis, training, multi-gpu distributed inference.