aditya gole

I'm building AIRA-SLM, a GPT-style small language model from scratch in PyTorch, implementing embeddings, attention, BPE tokenization, and autoregressive training loops by hand.

Alongside this research, I'm shipping Aira, an AI wardrobe recommendation engine with under-200 ms recommendation latency and embedding-based matching that exceeds a rule-based baseline by about 40%. I'm also architecting Costra, an AI electronics purchase decision engine that delivers citation-grounded recommendations.

At Thyssenkrupp Materials Services, I build backend services for a 50,000+ product catalogue and thousands of daily B2B transactions, reducing defects by 30%, PostgreSQL read latency by 40%, and release cycles from two hours to 20 minutes. Previously at Atek Payments, I built real-time ticketing and payment systems processing 5,000+ daily transactions with 99.9% uptime.