Manoj P. Saha — Agentic AI and AI systemshttps://pravakar.xyz/en-ushttps://pravakar.xyz/blog/ai-privacy/redact-transparent-privacy-filter-for-ai/https://pravakar.xyz/blog/ai-privacy/redact-transparent-privacy-filter-for-ai/Redact intercepts traffic between your machine and AI services, automatically anonymizing PII before it leaves your device — no workflow changes required.Sun, 10 May 2026 00:00:00 GMThttps://pravakar.xyz/blog/applied-ai/llm-assisted-threat-modeling/https://pravakar.xyz/blog/applied-ai/llm-assisted-threat-modeling/Using large language models to accelerate threat modeling sessions and identify attack surfaces in complex systems.Fri, 20 Mar 2026 00:00:00 GMThttps://pravakar.xyz/blog/ai-systems/lattice-efficient-in-memory-dnn-model-versioning/https://pravakar.xyz/blog/ai-systems/lattice-efficient-in-memory-dnn-model-versioning/A persistent memory library that makes DNN checkpoint writes faster by versioning models directly on non-volatile memory devices, bypassing the traditional storage bottleneck.Sun, 01 Jun 2025 00:00:00 GMThttps://pravakar.xyz/blog/data-systems/metadata-management-patent/https://pravakar.xyz/blog/data-systems/metadata-management-patent/A patented method for managing metadata in large block volumes by leveraging CXL-attached persistent memory devices for improved indexing performance.Tue, 01 Oct 2024 00:00:00 GMThttps://pravakar.xyz/blog/data-systems/moke-modular-kv-emulator/https://pravakar.xyz/blog/data-systems/moke-modular-kv-emulator/A QEMU-based KV-SSD emulator that accurately models device internals — data placement, garbage collection, and in-storage indexing — enabling realistic research without physical hardware.Sat, 01 Jul 2023 00:00:00 GMThttps://pravakar.xyz/blog/ai-systems/allocation-policies-hybrid-memory-systems/https://pravakar.xyz/blog/ai-systems/allocation-policies-hybrid-memory-systems/How the choice of memory allocation policy significantly impacts performance in hybrid DRAM + CXL memory systems, challenging the default DRAM-preferred approach.Tue, 20 Jun 2023 00:00:00 GMThttps://pravakar.xyz/blog/data-systems/leveraging-keys-kvssd/https://pravakar.xyz/blog/data-systems/leveraging-keys-kvssd/A scheme that exploits the hierarchical structure embedded in key names to improve I/O request handling performance for Key-Value and Object storage devices.Tue, 20 Jun 2023 00:00:00 GMThttps://pravakar.xyz/blog/data-systems/rhik-hash-based-indexing-kvssd/https://pravakar.xyz/blog/data-systems/rhik-hash-based-indexing-kvssd/A dynamically resizable hash-based index for KV-SSD devices that maintains high performance as data grows, solving the capacity and performance degradation problems of current designs.Tue, 20 Jun 2023 00:00:00 GMThttps://pravakar.xyz/blog/data-systems/kv-storage-optimization-patent/https://pravakar.xyz/blog/data-systems/kv-storage-optimization-patent/A granted US patent covering the RHIK resizable hash-based indexing technique that enables KV-SSDs to maintain performance at high occupancy without costly full-table rehashing.Sat, 01 Oct 2022 00:00:00 GMThttps://pravakar.xyz/blog/data-systems/kvssd-what-is-it-good-for/https://pravakar.xyz/blog/data-systems/kvssd-what-is-it-good-for/The first comprehensive performance characterization of Key-Value SSDs, revealing how their behavior differs from block SSDs and identifying workloads where they excel.Wed, 01 Dec 2021 00:00:00 GMThttps://pravakar.xyz/blog/data-systems/fine-grained-concurrency-kvssds/https://pravakar.xyz/blog/data-systems/fine-grained-concurrency-kvssds/Methods for managing concurrent access within KV-SSD devices to improve throughput and reduce tail latency under mixed workloads.Tue, 01 Jun 2021 00:00:00 GMThttps://pravakar.xyz/blog/ai-systems/performance-consistency-distributed-dl/https://pravakar.xyz/blog/ai-systems/performance-consistency-distributed-dl/A study of how system configurations, distribution topologies, and parameter server architectures affect the accuracy-throughput tradeoff in distributed deep learning training.Sun, 15 Nov 2020 00:00:00 GMT