Enhancing Venture Deal Discovery and Analysis with Artificial Intelligence.

Creating AI tools for:• Product Web Hierarchy categorization: Designed and trained AI model to determine the web hierarchy from the product’s title, and brand name. • Image de-duplication: Identify all copies of a target image in a database, including variations in size, perspective, orientation, color etc.

Exploring AI at the Edge, and Multi-modal AI • Controlling an Edge device using verbal commands – implemented a limited vocab. Natural Language Processing (NLP) model running on an iPhone. Real-time response time (~250ms) between the utterance and command recognition.• Multi-modal (images and voice) AI to reduce time spent on a home inspection by 40%. AI model combines a custom-trained CNN for images and LLM for language.

ML and AI algorithm design and modeling.• Audio anomaly detection – underwater sound classification, and identification of previously unheard sounds (anomalies), through feature extraction using deep neural nets (DNN) and clustering algorithms. • Position, Navigation and Timing (PNT) system and algorithm for local positioning where GPS is either unavailable, unreliable, or not preferred. Applications include autonomous driving, Vehicle to Everything (V2X) system, drone navigation.• Image classification, and object detection: classification and identification from image captures, using convolutional neural net (CNN) architectures. Identify and extract lettering on objects (like tires) from the image.• Visual product categorization - to recognize food products that do not meet quality standards, from an image. Trained a CNN model using transfer learning - achieved an accuracy of 95%.• Edge Solar monitoring solution that uses machine learning to predict the solar plant output under the prevalent environmental conditions, and flags if the predicted output and actual output differ. A distributed AI solution running on a network of IoT devices, and the Cloud.

Helping farmers in India to use the Sun for one more thing - electricity, for both their irrigation needs and power at their homes.

In 2012, two years before Alexa, my team designed and demonstrated a novel, a first of its kind, processor IC for Machine Learning / AI algorithms. Designed as an accelerator to an ARM CPU, the architecture was optimized for data transfer between the multiple Compute Engines (CE) using highly efficient memory structures, DMA engines and handshaking algorithms implemented in hardware. These semi-autonomous CE blocks could be organized into different topologies including SIMD, pipelined etc. This was a DARPA funded project.

Extremely low power analog signal processing technology. Explored applications including hearing aids, error correction (LDPC), machine learning algorithms.

Specifications, Architecture and Design of the Mixed Signal Power Conversion ASIC for a solar micro-inverter

Excelled in various technology and application areas, and in various capacities – team management, signal processing and algorithm development, IC architecture, and design.Innovative firsts (multiple US patents) in Sigma-Delta converters (ADCs and DACs), Digital Signal Processing (FIR, IIR, DSPs), IC Communication, Power Control algorithms (PFC, buck, AC-DC)Team lead and technical lead for an innovative Delta Sigma ADC family for the industrial markets - CS5560 (Finalist (1 of 4) in the "Top Analog IC" category for the 2007 EDN Innovation Awards)