Aryan Dhanuka

I’m an AI engineering-focused developer who builds reliable, low-latency machine learning and LLM systems designed for production realities—testing rigor, measurable performance, and clean interfaces.

In my open source work with sktime, I developed scenario-based tests for detector estimators and implemented DetectorUnivariateSimple with comprehensive smoke tests to improve reliability. I also enhanced the TimeBinner transformer with advanced test parameter sets, improving robustness for time-series binning and addressing critical edge-case failures.

As an AI Engineering Intern at LearnNex (Powered by Wipro), I built a production-grade multi-domain LLM system with domain-aware routing and modular agents. I developed a hybrid RAG pipeline (BM25 + dense embeddings), created a document intelligence workflow (PDF ingestion, chunking, Map-Reduce summarization), and deployed scalable FastAPI inference APIs with streaming and low-latency response handling.

My projects reflect the same mindset: turning prototypes into dependable systems. I shipped an “infertrack” pip-installable LLM observability library with local-first token and cost tracking, engineered a hybrid multimodal manufacturing quality failure detector with uncertainty-aware HITL, and built a multi-domain intelligent assistance system with fast streaming and confidence-threshold routing—replacing costly embeddings with local Hugging Face models to cut latency and per-token cost.