Sthiti Deka Email & Phone Number

Santa Clara, Ca, Us

Santa Clara, California, Us

- Analyze AI workloads for efficient inference on NPU- Analyze accuracy impact of model quantization

AI/ML Research - Augment health data sets using synthetic EHR data generation - Build agents for symptom checking, risk analysis, wellness recommendation using a combination of prompt engineering, RAG and fine tuning LLMs. Frameworks: Langchain, LlamaIndexLLMs: gpt-3.5-turbo, text-bisonRetrievers: FAISS, GPTIndex, BM25

Palo Alto, Ca, Us

Performance Modeling, Benchmarking, Characterization of ML Workloads:- Developed and tuned computational models to generate performance projections of inference & training graphs for target architectures - Developed benchmark suite for architecture evaluation, analysis and scaling- Characterized workloads as compute, capacity, bandwidth or network bound - Influenced design decisions of future hardware generationsProduct Architect and Tech Lead:- Concept, architect and design lead of a performance profiling and tuning product with features that help compare and contrast runs with parameter sweeps, collates results, tracks experiments, highlights bottlenecks and makes tuning recommendations. Analysis results are visualized in a dashboard with capability to interact with the data.- Responsible for roadmap, planning and execution - Responsible for staffing and growing talent- Responsible for managing and mentoring team and providing technical guidance

Santa Clara, California, Us

Visual Processor Graphics: - Implemented features and optimizations in the compiler for Metal on OSX - Implemented OpenCL API in the driver for GPGPU on Windows and Mac OS - Implemented OpenCL-C kernel language features and optimizations in an LLVM based compiler - Debugged applications in user-mode and kernel-mode layers in simulated and real hardware environment - Enabled non-graphics workloads via OpenCL on GPGPU - Implemented power-performance optimizations in the driver for integrated graphics - Profiled and analyzed performance bottlenecks in compute benchmarks, 3D games, image & video editing studios and tuned via driver and compiler optimizations - Performed competitive benchmarking and drove optimizations across the stack

Santa Clara, California, Us

Multimedia Engineering, Graphics Processor Engineering Group- Profiled DivX encoder, decoder, MPEG-2, Blu ray encoder plug-ins in Sony Vegas Platinum Movie Studio, Adobe Encore CS4 to identify performance bottlenecks- Benchmarked media applications implementing compression standards (h.264, blu-ray, mpeg-2), media containers(Matroska, VLC), media interfaces(DXVA 10.0)- Measured differences in latency and bandwidth limits in discrete and integrated graphics cards - Studied and characterized shared memory architecture of Tigris integrated graphics platform via benchmark applications- Participated in a feasibility study of CPU-GPU load balancing for accelerated performance of multimedia plugins- Tested AMD’s implementation of OpenCL specifications, contrasted performance with CAL, Brook+

Amsterdam, Noord-Holland, Nl

- Implemented data access interfaces for storage and retrieval of medical images to and from media, network nodes and server systems- Explored the challenges and opportunities of a Content Based Image Retrieval system for medical images

Master'S Degree, Electrical & Computer Engineering

Bachelor'S Degree, Electrical And Electronics Engineering