Technical Articles

• A Very Short Overview of Development of NLP in Recent Times

  This essay provides a brief overview of the evolution of Natural Language Processing (NLP) technologies, with a focus on recent advancements.

• The Evolutionary Journey of Search Engines: Architecture and Evaluations

  Delve into a concise exploration tracing the history and evolution of modern search engines. This document offers insights into their underlying architecture and the pivotal methods used for their evaluation.

• LLMs and the Future of Search

  Explore how Large Language Models (LLMs) are poised to revolutionize search, building upon the sophisticated AI systems and complex engineering that power today's search engines, from web crawling and indexing to query understanding and ranking. This document examines the evolution of search and the potential impact of LLMs on its future.

• The Dawn of LLMs: A Visionary Perspective on the Future of Search

  Engage with a thought-provoking opinion piece, tailored for a technical audience, that envisions the transformative potential of Large Language Models in reshaping the future landscape of search.

• Licensing Code: A Primer for Software Engineers

  A brief note on licensing softwares.

• Ethical Considerations and Regulations in Artificial Intelligence

  A brief discussion on AI ethics and necessity of regulations.

• Summary of US government's executive order on AI

  It is a quick read to understand the essence of President Biden's executive order on AI.

• A Concise Review of Reinforcement Learning Methods: Foundations, Deep RL, & LLM-centric Approaches

  A brief overview of reinforcement learning (RL) methods, focusing on the foundations, deep RL techniques, and their application within the context of Large Language Models (LLMs).

• Notes on Model Distillation

  A review of the literature on model distillation, exploring algorithms and applications of this important method for model compression and efficiency.

• The Evolution Beyond Search – From Information Retrieval to Autonomous Execution

  Part 2 of the series on LLMs and the Future of Search (see also Part 1). This article explores the progression of search beyond simple information retrieval, examining the potential for autonomous execution and the role of Large Language Models.

• Biological Intelligence and Machine Learning: A Comparative Analysis of OpenAI and Anthropic Approaches

  This analysis delves into the contrasting model development strategies employed by OpenAI and Anthropic, two of the most influential AI research organizations.

• Agents and Agentic AI in the Era of Large Language Models

  This essay provides a comprehensive overview of agentic AI, explaining what agents are, how Large Language Models facilitate their creation through capabilities like planning and tool use, and what constitutes an effective agent platform. It describes suitable LLMs, details existing frameworks (e.g., LangChain, AutoGen) and managed services (e.g., OpenAI Assistants), comparing their features, pros, and cons, while also covering implementation specifics, workflow design, and cost factors to offer readers a solid understanding of this rapidly evolving space.

• Search 2025: From Information Access to Intelligent Execution

  In this essay, I explore how LLMs are transforming search from a user-facing interface into invisible infrastructure that enables real-time reasoning, retrieval, and execution within intelligent systems..

• The Rise of Reasoning Models: A Revival of Planning in AI

  From prediction to deliberation — how modern AI is rediscovering the value of structured reasoning.

• Data Engines for Autonomous Vehicles: A Reference Architecture

  This paper presents a reference architecture for production data engines in autonomous vehicle systems—the closed-loop infrastructure that transforms raw sensor data for training ML models.

• Architectural Advances in AI Through a Systems Lens

  This paper examines most significant developments across the full AI stack—from model architecture and training paradigms to compiler infrastructure, runtime systems, accelerators, interconnect, and scientific applications—and shows that the practical impact of modern AI architectures is constrained less by model design than by the maturity of the execution stack beneath them.

• Enterprise AI Outlook 2026 and Beyond: Applications, Inference Optimization, and Infrastructure

  This white paper examines the transition of AI from model-centric experimentation to enterprise-scale application and system deployment. It analyzes trends across training, inference optimization, agentic application design, hardware infrastructure, data strategy, and evaluation, and argues that beyond 2026 the primary constraints on AI impact are economic, architectural, and operational rather than model capability alone.

• GPU Cluster Infrastructure Outlook 2026 and Beyond: Power, Compute, and Strategic Constraints

  This white paper analyzes the global GPU cluster infrastructure inflection driven by AI-scale workloads. It examines market expansion, hyperscaler capex, geographic concentration, accelerator competition, power and cooling constraints, memory and interconnect limits, and supply-chain dynamics, arguing that through 2028 the binding constraints on AI progress are infrastructure readiness, energy availability, and system-level reliability rather than silicon performance alone.

• The Adobe Document Cloud Disruption Thesis: PDF Wins, But Does Adobe? (2025–2035)

  This analysis examines the paradox facing Adobe's $3.18B Document Cloud business: PDF succeeds as the universal document format while Adobe faces disintermediation at every value-creating layer. It maps programmatic generation bypass, e-signature commoditization, and AI-native document intelligence capture by startups like Reducto and Unstructured, arguing that 35–50% of Document Cloud's addressable market is vulnerable by 2035—unless Adobe executes an aggressive AI-native pivot through acquisition or "Firefly for Documents" capabilities.

• The Transformation of Ad Engines: LLMs, Foundation Models, and Agentic Systems (2026+)

  This analysis examines digital advertising's most significant transformation since programmatic buying, as LLMs, foundation models, and autonomous agents reshape every component of the ad stack. It details Meta's GEM achieving 5% conversion lifts, Google's AI Overviews reaching 1.5B users with integrated ads, and the collapse of organic CTR from 7.3% to 2.6%. The report argues that while AI enables unprecedented efficiency, a countervailing consumer reckoning looms—58% replacing search with AI, 50% limiting social media usage, and only one-third viewing AI advertising positively—forcing platforms to balance technological capability against eroding consumer trust.

• State of AI Silicon 2026: The Definitive Technical Atlas

  This technical report maps the GPU design landscape through the engineering constraints that now matter more than raw compute: memory bandwidth, packaging capacity, and software ecosystem maturity. It compares NVIDIA's dual-die Blackwell (208B transistors, 8 TB/s), AMD's chiplet MI300X (33% TCO advantage for memory-bound inference), Cerebras's wafer-scale 21 PB/s on-chip bandwidth, and China's Ascend 910C reaching 60% H100 performance through system-scale brute force. The analysis covers supply chain bottlenecks (HBM sold out through 2026, CoWoS at 60% NVIDIA allocation), the CUDA moat (6M developers vs. ROCm/CANN fragmentation), and market dynamics as inference spending overtakes training—projecting value migration from NVIDIA's 86–94% training monopoly toward a fragmented $254B inference market by 2030.