Tech lead for production GenAI platforms across finance automation, realtime voice AI, and regulatory search. Designed systems for production reliability, latency, secure deployment, and operational handoff.

Stack: Python, FastAPI, Google ADK, Vertex AI, OpenAI, Milvus, Redis, MongoDB, Docker, GCP.

HCL Projects

IFRS — Invoice anomaly detection and FP&A variance analysis with Gemini/Google ADK, SAP data validation, and Vertex AI deployment.
AACA — Production realtime voice AI with Twilio, OpenAI Realtime API, Milvus RAG, structured tool calling, and live human handoff.
SEBiSearch — Offline regulatory RAG with hybrid Jina/BM25 retrieval, ModernBERT reranking, citation-constrained generation, and Redis/FAISS caching.

Shipped edge vision and accelerator software across computer vision, BEV segmentation, and deep learning kernel optimization. Built production C++ pipelines, trained vision models, and mapped deep learning operators to accelerator kernels.

PathPartner / KPIT Projects

Image Stitching — ONNX and C++ feature detection/matching pipeline on Qualcomm, reducing CPU runtime by 75%.
CES — SOLOv2 training, OpenCV BEV blending, multi-model segmentation/classification, and LIME visualization.
Deep Learning Kernels — PyTorch/TensorFlow to TPC kernels in C++, with ONNX and Netron delivery.

Built Firebase telemetry and ML data-processing foundations during an engineering internship, including on-demand preprocessing and Baum-Welch HMM parameter and state estimation.

AgVerse Project

AgVerse Analytics — Firebase telemetry, on-demand preprocessing, and Baum-Welch HMM estimation.