Hardware (AI Silicon) Internship at NVIDIA
TextEntity Ner Classification
I architected the domain-specific training data pipeline for a custom Multi-Agent LLM Framework built on LangChain's ReAct architecture. To train the AI agents to accurately understand and query complex semiconductor metrics, I engineered recursive Python and SQL logic that extracted, interpolated, and reconciled deeply nested YAML-based Timing Configurations into structured, ground-truth datasets. By transforming these highly technical files into optimized Apache Parquet formats queried via DuckDB, I provided the high-fidelity, low-latency data environment required for the LLM to learn context, perform dynamic statistical analysis, and generate accurate responses across diverse process-voltage-temperature corners.
I architected the domain-specific training data pipeline for a custom Multi-Agent LLM Framework built on LangChain's ReAct architecture. To train the AI agents to accurately understand and query complex semiconductor metrics, I engineered recursive Python and SQL logic that extracted, interpolated, and reconciled deeply nested YAML-based Timing Configurations into structured, ground-truth datasets. By transforming these highly technical files into optimized Apache Parquet formats queried via DuckDB, I provided the high-fidelity, low-latency data environment required for the LLM to learn context, perform dynamic statistical analysis, and generate accurate responses across diverse process-voltage-temperature corners.
2025 - 2025