Nikkat Afrin

I'm a Data & AI Platform Engineer at Peblink / World Literacy Research Center, where I built a speech-AI reading assessment pipeline that reaches 93% pronunciation-detection accuracy and lets teachers flag dyslexia-risk reading patterns in seconds. It took review time from over an hour to under five minutes.

My work sits where machine learning meets real infrastructure: not notebooks that prove a point, but systems that keep running. Serverless AWS pipelines with S3, SQS and Dockerized Lambda. RAG assistants grounded in real documents. Dimensional warehouses that preserve history. Across five roles in education, logistics and admissions I've kept coming back to the same problem, people drowning in manual work that data should have handled. So I automated Symplicity ingestion and gave back 40+ staff hours a week, co-built FELIX, a RAG chatbot answering student Canvas and career questions, and turned literacy research into product requirements that earned a meeting in Washington, D.C. with Congressman Brad Schneider.

On the research side I'm first author of SigmaCam (IEEE IJCNN 2025), extracting exact decision boundaries for smooth-activation networks that prior exact methods couldn't handle, cutting runtime ~93% while holding ~99.8% accuracy. It won Best Research Project at Yeshiva's CSE Research Forum. Throughout all of it I care a lot about honest reporting: showing the baseline, naming what didn't work, and refusing to dress up a number that doesn't mean anything.