Data Science Intern
Shipped a multilingual AI voice agent for telecom clients and led front-to-back migration work on enterprise AI assistants. Focus on low-latency retrieval, multilingual parsing, and reliability under real customer traffic.
01
The brief
NCINGA's telecom AI was scaling up across Sri Lanka and needed to serve customers natively in English, Sinhala, and Tamil, without forcing anyone to switch channels. My summer was to ship the multilingual voice side of that, and then take on a wider role across the enterprise AI assistants the team was rolling out.
02
What I built
The core piece was a multilingual AI voice agent that could understand and respond in real time across the three languages. The hard part was phone number normalization, because customers say numbers in at least fifteen different spoken variations across the three. I wrote a Python parser that handled the variations and validated to 99 percent accuracy before the agent acted on anything.
03
Taking on the assistants
The second half of the internship I led a full-stack migration on the enterprise AI assistants, including a branding refresh and moving the internal knowledge search behind a lower-latency API layer. Query resolution time for the internal teams came down by 40 percent.
04
What I took from it
Working with real customers across three languages teaches you fast that latency and language parsing are not secondary concerns, they are the product. I left NCINGA with a strong respect for how the unglamorous reliability work actually shows up in user experience.
Gallery
