SanDisk India
Machine Learning Engineer · via Magnit Global
- Built SwiftECO — BiLSTM + Multi-Head Attention model for multi-corner STA timing prediction with asymmetric loss and cross-design transfer learning: R² = 0.91, MAE ~6.7 ps, cutting analysis turnaround from 24+ hours to 47 minutes
- Built NLP/RAG pipeline over 2000+ pages of technical documentation (LangChain, FAISS, AWS Bedrock) — natural-language querying over engineering knowledge bases
- Designed scalable pipelines for 10+ GB PrimeTime reports — automated ETL, statistical profiling, and MLflow-tracked validation for reproducible model governance
- Shipped RTL Lint & Stream IP agents automating Verilog cleanup with designer-in-the-loop review; improved STA timing consistency by 55%
PyTorch
LangChain
FAISS
AWS Bedrock
MLflow
