Tejas Narayanan

Developed an end-to-end Python/C++ pipeline to process, enrich, and stream real-time trading position data. Parallelized column enrichment and queries using multi-threaded Kafka consumers, achieving an 8x processing speedup. Coordinated with traders to design and launch custom filter layers in production.

Optimized an Apache Kafka integration pipeline receiving real-time market trade signals. Prevented major lag spikes under extreme load spikes by building asynchronous batch commits to a Cassandra database cluster, delivering a 2,200x speed improvement under stress tests.

Designed a reinforcement learning pipeline for adaptive computer vision data selection using OpenAI Gym and PyTorch. Built a Neural Data Filter (NDF) to dynamically evaluate and select informative training images, improving downstream model accuracy by 27% on production driving datasets.

Developed convolutional neural networks in PyTorch to predict regional income distributions from high-resolution satellite imagery. Achieved the highest validation accuracy in the Open division for the Fall 2020 project cycle.

Created a 2D ocean navigation simulator in Pygame to model ocean currents and obstacles. Authored path-planning routines using Voronoi diagrams and Dijkstra/A* search to autonomously navigate an ocean-bound research vessel mapping Palau's coral reefs.