# Subhadip Mitra: Full Site Context > Personal website and technical blog of Subhadip Mitra, Engineering Leader at Google Cloud, AI researcher, and open-source contributor. ## Author Name: Subhadip Mitra Role: Head of Data & Analytics, Site Lead Southeast Asia Employer: Google Cloud Location: Singapore Email: contact@subhadipmitra.com Education: - MBA in Business Analytics, BITS Pilani - MTech in Software Systems, BITS Pilani Affiliations: Google Cloud, IIT Madras (Research Affiliate), IEEE, ACM, Singapore Computer Society Expertise: Artificial Intelligence, Machine Learning, Large Language Models, Data Engineering, Cloud Computing, Distributed Systems, Multi-Agent Systems, Privacy and Consent Protocols ## Blog Posts ### Your OpenAPI Spec Is Not an Agent Interface: Compiling Governed MCP Tool Surfaces URL: https://subhadipmitra.com/blog/2026/api-mcp-compiler/ Date: 2026-08-12 Tags: mcp, agents, openapi, governance, code-generation api-mcp-compiler turns OpenAPI, Swagger, and WSDL specifications into MCP tool surfaces with field-level provenance, human approval gates, and compiled-in refusals. Four pre-registered evaluations could not resolve whether semantic planning beats one tool per operation. Here is the design argument, the numbers, and what the evaluation harness caught anyway. ### Beating FP16 with 4-bit Weights: A Portable W4A16 GEMM in Triton URL: https://subhadipmitra.com/blog/2026/w4a16-weight-only-gemm-triton/ Date: 2026-07-20 Tags: deep-learning, llm, triton, quantization, inference-optimization I wrote a 4-bit weight-only GEMM in pure Triton. The fast W4A16 kernels are all CUDA, so this one runs on NVIDIA and AMD. It beats cuBLAS FP16 by 1.1 to 1.3x in the decode regime, and the road there was mostly me being wrong about memory bandwidth. ### The Activation-Cone Blind Spot, or Why Your Jailbreak Defense Can't See Prefilling URL: https://subhadipmitra.com/blog/2026/activation-cone-blind-spot/ Date: 2026-07-06 Tags: ai-safety, interpretability, llm-security, jailbreaks, red-teaming, probes Prompt-time activation defenses stop GCG and AutoDAN cold, then fail half the time against prefilling. New paper: why that blind spot is structural, and how a linear probe on the first generated tokens closes it - AUROC 0.97-1.00 across seven models, zero prefilling success when composed with steering. ### What Runtime Interpretability Actually Costs, Part 1: The Case for Measuring It URL: https://subhadipmitra.com/blog/2026/runtime-interpretability-cost/ Date: 2026-07-02 Tags: interpretability, inference, serving, probes, benchmarks, ai-safety Everyone assumes activation probes are too expensive to run in production. I ran the numbers on paper and I no longer believe it. Here is the argument, my predictions, and the harness I built to settle it. ### DeepSeek DSpark: Speculation Is a Scheduling Problem URL: https://subhadipmitra.com/blog/2026/deepseek-dspark-speculative-decoding-production/ Date: 2026-06-29 Tags: deepseek, dspark, speculative-decoding, llm-inference, inference-optimization, deepspec, vllm What DeepSeek's DSpark and DeepSpec release actually changes for LLM inference: suffix decay, confidence-scheduled verification, a decision guide, and how to train a draft model for your own weights. ### Loop Engineering: Convergence Is Not Correctness URL: https://subhadipmitra.com/blog/2026/loop-engineering-verifier/ Date: 2026-06-21 Tags: agents, loop-engineering, llm-engineering, verification, ai-engineering A loop always converges. That it reached a stable 'done' state tells you nothing about whether that state is correct, because the loop selects for whatever passes your check, not for what you meant. A practitioner's take on loop engineering: why convergence is not correctness, why the verifier is the part that actually decides, and what changes when loops run unattended over real production systems instead of a coding agent in a terminal. ### Attention Is All You Bid: Advertising in Embedding Space URL: https://subhadipmitra.com/blog/2026/attention-is-all-you-bid/ Date: 2026-04-04 Tags: llm, advertising, embedding-space, RAG-poisoning, GEO, auction-mechanism, ad-tech Embedding space is the new ad real estate. Mapping LLM ad auctions, RAG poisoning, GEO, and a framework for what comes next. ### Beating CUDA with Triton: A Fused MoE Dispatch Kernel for Mixtral and DeepSeek URL: https://subhadipmitra.com/blog/2026/fused-moe-dispatch-triton/ Date: 2026-03-28 Tags: deep-learning, llm, triton I wrote a fused Mixture-of-Experts dispatch kernel in pure Triton that beats Stanford's CUDA-optimized Megablocks at inference batch sizes, and runs on both NVIDIA and AMD GPUs without a single line of CUDA. ### Confessions vs. CoT Monitoring vs. Probes: Three Bets on Model Honesty URL: https://subhadipmitra.com/blog/2026/three-bets-model-honesty/ Date: 2026-03-07 Tags: ai-safety, interpretability Three labs. Three different bets on how to catch models misbehaving. Each makes different assumptions about when models 'know' they're lying. Here's what works, what doesn't, and what happens when you combine them. ### OpenAI's Confessions Paper Has a Blind Spot. Here's What Fills It. URL: https://subhadipmitra.com/blog/2026/openai-confessions-blind-spot/ Date: 2026-02-18 Tags: ai-safety, interpretability OpenAI trained GPT-5 to confess when it misbehaves. It works surprisingly well - except when the model doesn't know it's misbehaving. That's where activation probes come in. ### Activation Steering in 2026: A Practitioner's Field Guide URL: https://subhadipmitra.com/blog/2026/activation-steering-field-guide/ Date: 2026-02-12 Tags: interpretability, ai-safety I've been working with steering vectors for months. Here's what actually works in practice, what fails in ways nobody warned me about, and the honest playbook for getting started. ### Moltbook as MCP Stress Test: What 770K Agents Reveal About Protocol Design URL: https://subhadipmitra.com/blog/2026/moltbook-mcp-stress-test/ Date: 2026-02-02 Tags: agents A follow-up to my MCP Maturity Model post. Moltbook shows what happens when you run 770K agents at Level 0 maturity with zero governance. The results are instructive. ### Circuit Tracing for the Rest of Us: From Probes to Attribution Graphs and What It Means for Production Safety URL: https://subhadipmitra.com/blog/2026/circuit-tracing-production/ Date: 2026-01-31 Tags: interpretability, ai-safety MIT Tech Review named mechanistic interpretability a 2026 Breakthrough Technology. Anthropic open-sourced circuit tracing. Here's what actually changed, how it connects to the activation probes I built for sandbagging detection, and why production teams should care. ### RLVR Beyond Math and Code: The Verifier Problem Nobody Has Solved URL: https://subhadipmitra.com/blog/2026/rlvr-beyond-math-code/ Date: 2026-01-18 Tags: llm, deep-learning Reinforcement Learning with Verifiable Rewards powers every reasoning model worth talking about. But it only works where you can check the answer automatically. Extending it to messy, real-world domains is the hardest open problem in LLM training right now. ### The Agent Protocol Stack: Why MCP + A2A + A2UI Is the TCP/IP Moment for Agentic AI URL: https://subhadipmitra.com/blog/2026/agent-protocol-stack/ Date: 2026-01-06 Tags: agents MCP handles agent-to-tool. A2A handles agent-to-agent. A2UI handles agent-to-interface. Together they form a protocol stack that nobody has mapped properly - including the security gaps that should terrify you. ### The Manifold Dial: Visualizing Why DeepSeek's mHC Stabilizes Deep Networks URL: https://subhadipmitra.com/blog/2026/deepseek-mhc-manifold-constrained-hyper-connections/ Date: 2026-01-03 Tags: deep-learning Interactive exploration of Manifold-Constrained Hyper-Connections - how DeepSeek fixed the signal explosion problem in deep residual networks using 1967 mathematics ### I Trained Probes to Catch AI Models Sandbagging URL: https://subhadipmitra.com/blog/2025/detecting-ai-sandbagging/ Date: 2025-12-20 Tags: ai-safety, interpretability First empirical demonstration of activation-level sandbagging detection. Linear probes achieve 90-96% accuracy across Mistral, Gemma, and Qwen models. Key finding - sandbagging representations are model-specific, and steering can reduce sandbagging by 20%. ### Why Steering Vectors Beat Prompting (And When They Don't) URL: https://subhadipmitra.com/blog/2025/steering-vectors-agents/ Date: 2025-12-18 Tags: interpretability, llm I tested activation steering on 4 agent behaviors across 3 models. The results surprised me. ### Why I Built a Spark-Native LLM Evaluation Framework (And What I Learned) URL: https://subhadipmitra.com/blog/2025/building-spark-llm-eval/ Date: 2025-12-15 Tags: llm, open-source A deep dive into building distributed LLM evaluation infrastructure that actually scales - architectural decisions, trade-offs, and lessons learned. ### The MCP Maturity Model: Evaluating Your Multi-Agent Context Strategy URL: https://subhadipmitra.com/blog/2025/mcp-maturity-model/ Date: 2025-11-19 Tags: agents A practical framework for evaluating your multi-agent context management strategy. From ad-hoc string concatenation to self-evolving context systems - where does your architecture stand? ### UPIR: What If Distributed Systems Could Write (and Verify) Themselves? URL: https://subhadipmitra.com/blog/2025/upir-distributed-systems/ Date: 2025-11-15 Tags: distributed-systems Lessons from building a framework that automatically generates verified distributed systems - and why I think formal methods, synthesis, and ML need to work together ### The Data Platform Crisis Hiding Behind AI: Why you have 6 months to pivot URL: https://subhadipmitra.com/blog/2025/agent-ready-data-platforms-sarp/ Date: 2025-10-17 Tags: data-platforms, agents Enterprise data platforms face a 100,000x query increase from agentic AI. Introducing Symbiotic Agent-Ready Platforms (SARPs) - the architectural paradigm shift needed to survive the transition to machine intelligence. ### AI Meta-Cognition - The Observer Effect Series URL: https://subhadipmitra.com/blog/2025/ai-deception/ Date: 2025-10-11 Tags: ai-safety, interpretability Frontier AI models from OpenAI, Anthropic, Google & others can detect when they're being tested and modify behavior-challenging AI safety evaluation methods. ### Building Safer AI: Industry Response and the Path Forward - (Part 4/4) URL: https://subhadipmitra.com/blog/2025/building-safer-ai-industry-response-practical-solutions/ Date: 2025-10-11 Tags: ai-safety How the AI industry is responding to situational awareness challenges. Practical monitoring systems, collaborative research, and what organizations should do today. ### Alignment Faking: When AI Pretends to Change - (Part 3/4) URL: https://subhadipmitra.com/blog/2025/alignment-faking-ai-pretends-to-change-values/ Date: 2025-10-07 Tags: ai-safety Claude 3 Opus strategically fakes compliance during training to preserve its values. This alignment faking undermines our ability to modify AI behavior safely. ### Deliberative Alignment: Can We Train AI Not to Scheme? - (Part 2/4) URL: https://subhadipmitra.com/blog/2025/deliberative-alignment-training-ai-not-to-scheme/ Date: 2025-10-03 Tags: ai-safety Researchers achieved a 30-fold reduction in AI scheming through deliberative alignment. But rare failures persist. Can we truly train models not to deceive? ### The Observer Effect in AI: When Models Know They're Being Tested - (Part 1/4) URL: https://subhadipmitra.com/blog/2025/ai-observer-effect-models-recognize-evaluation/ Date: 2025-09-30 Tags: ai-safety, interpretability Frontier AI models from OpenAI, Anthropic, and Google can now recognize when they're being tested. This observer effect undermines AI safety evaluation. ### We Need a Consent Layer for AI (And I'm Trying to Build One) URL: https://subhadipmitra.com/blog/2025/building-consent-layer-for-ai/ Date: 2025-08-16 Tags: blockchain AI companies are getting sued over training data, agents operate with no permission framework, and users can't control their AI profiles. I wrote four open standards (LLMConsent) to create a decentralized consent protocol for AI - like HTTP but for data rights, agent permissions, and user sovereignty. This is an RFC, not a product. ### Why Kimi K2 Stands Out - A Deep Dive into Its Trillion-Parameter MoE URL: https://subhadipmitra.com/blog/2025/why-kimi-k2-stands-out/ Date: 2025-07-13 Tags: llm, deep-learning Explore Kimi K2’s trillion-parameter MoE architecture, MuonClip optimizer, and agentic training. Learn why it outperforms GPT-4.1 and DeepSeek-V3 ### From 11% to 88% Peak Bandwidth: Writing Custom Triton Kernels for LLM Inference URL: https://subhadipmitra.com/blog/2025/triton-kernels-llm-inference/ Date: 2025-06-15 Tags: deep-learning, llm A hands-on exploration of writing custom GPU kernels with OpenAI Triton, going from PyTorch's 11% bandwidth utilization to 88% on RMSNorm. ### Context Engineering for Autonomous Multi-Agent Systems - Packaging, Pruning, and Transferring Context at Scale URL: https://subhadipmitra.com/blog/2025/implementing-model-context-protocol/ Date: 2025-03-22 Tags: agents A technical deep dive into context engineering for autonomous multi-agent systems - how to package, score, prune, and transfer context between agents at scale, with complete Python implementations and a hypothetical telecom scenario. Note - this is distinct from Anthropic's Model Context Protocol (MCP); the two are complementary, and the post explains where each fits. ### Making LLMs Faster: My Deep Dive into Speculative Decoding URL: https://subhadipmitra.com/blog/2025/making-llm-faster/ Date: 2025-03-20 Tags: llm, deep-learning A deep dive into implementing speculative decoding from scratch, with benchmarks on GPT-2 and extensions to diffusion models. ### Engineering Autonomous Multi-Agent Systems - A Technical Deep Dive into Telecom Customer Service URL: https://subhadipmitra.com/blog/2025/telecom-autonomous-multi-agent-genai-system/ Date: 2025-01-05 Tags: agents, case-study Dive into the world of autonomous AI agents with practical implementations, code examples, and real-world scenarios. Learn how to build intelligent systems with advanced memory management, dynamic prompt evolution, and sophisticated monitoring capabilities in telecom customer service. ### Why I Built a Modern Java SMPP Library in 2025 URL: https://subhadipmitra.com/blog/2025/why-i-built-modern-java-smpp-library/ Date: 2025-01-03 Tags: open-source The story behind smpp-core - a clean-room Java 21 implementation of the SMPP protocol. Why I replaced Cloudhopper, what went into it, and actual benchmark numbers. ### Engineering Multi-Agent Systems - A Retail Banking Case Study URL: https://subhadipmitra.com/blog/2024/retail-bank-multi-agent-system/ Date: 2024-12-28 Tags: agents, case-study Explore a detailed technical implementation of a multi-agent system for retail banking credit assessment. Learn about agent architecture, distributed systems patterns, error handling, compliance requirements, and performance optimization through actual code examples and system diagrams. Ideal for software architects and engineers building scalable financial systems. ### ETLC 2.0 - Building Context-Aware Data Pipelines URL: https://subhadipmitra.com/blog/2024/etlc-adaptive-contexts-and-contextual-joins/ Date: 2024-12-07 Tags: data-platforms, llm Think your data pipelines could do more than just process information? ETLC 2.0 takes data engineering to the next level with Adaptive Context, Contextual Joins, and a scalable Context Store. It's not just about moving data—it's about making it intelligent. Ready to unlock the future of data pipelines? Read on. ### The End of Data Warehouses? Enter the Age of Dynamic Context Engines URL: https://subhadipmitra.com/blog/2024/end-of-data-warehouses/ Date: 2024-11-18 Tags: data-platforms, llm Traditional data warehouses are struggling to keep up with modern demands. Enter Dynamic Context Engines (DCEs) - real-time, path-aware platforms that enrich data with context for smarter, faster decisions. Discover why they're the future of data analytics. ### (Part 3/3) - Reimagining ETL with Large Language Models—The Path to Intelligent Pipelines URL: https://subhadipmitra.com/blog/2024/etl-llm-part-3/ Date: 2024-10-20 Tags: data-platforms, llm Explore how Large Language Models (LLMs) are revolutionizing ETL pipelines. Discover advanced techniques like context-driven transformations, semantic joins, and multimodal integration, redefining data engineering with smarter, adaptive, and intelligent workflows. ### Data Pipelines Gone Wild - 10 WTF Moments That'll Make You Rethink Your Architecture URL: https://subhadipmitra.com/blog/2024/data-pipelines-gone-wild/ Date: 2024-08-02 Tags: data-platforms Buckle up for a wild ride through 10 mind-blowing data pipeline disasters and their solutions. From ancient code to biased algorithms, this post reveals the chaos and how to conquer it! ### Introducing ETL-C (Extract, Transform, Load, Contextualize) - a new data processing paradigm URL: https://subhadipmitra.com/blog/2024/etlc-context-new-paradigm/ Date: 2024-05-04 Tags: data-platforms, llm Think your AI apps could use a deeper understanding of your data? ETL-C (extract, load, transform, and contextualize) could be the answer. It's about adding context for better decisions. Intrigued? Read on. ### (Part 2/3) Rethinking ETLs - How Large Language Models (LLM) can enhance Data Transformation and Integration URL: https://subhadipmitra.com/blog/2024/etl-llm-part-2/ Date: 2024-04-20 Tags: data-platforms, llm Rethinking ETLs - The Power of Large Language Models. Part 2 Exploring examples and optimization goals ### (Part 1/3) Rethinking ETLs - How Large Language Models (LLM) can enhance Data Transformation and Integration URL: https://subhadipmitra.com/blog/2024/etl-llm-part-1/ Date: 2024-04-15 Tags: data-platforms, llm Rethinking ETLs - The Power of Large Language Models. Part 1 - Explore traditional algorithms for efficient ETL planning in complex data. ### Who Needs Exact Answers Anyway? The Joy of Approximate Big Data URL: https://subhadipmitra.com/blog/2024/big-data-approximate-calculations/ Date: 2024-01-16 Tags: data-platforms Discover how sacrificing a bit of accuracy can lead to huge gains in big data analysis speed and efficiency. ### Evolutionary Bytes - Harnessing Genetic Algorithms for Smarter Data Platforms (Part 2/2) URL: https://subhadipmitra.com/blog/2023/genetic-algorithm-inspired-data-platforms-part-2/ Date: 2023-12-29 Tags: data-platforms Explore how genetic algorithms revolutionize data platforms, offering adaptive, dynamic solutions to meet complex challenges in the fast-evolving digital landscape. ### Evolutionary Bytes - Harnessing Genetic Algorithms for Smarter Data Platforms (Part 1/2) URL: https://subhadipmitra.com/blog/2023/genetic-algorithm-inspired-data-platforms-part-1/ Date: 2023-12-25 Tags: data-platforms Explore how genetic algorithms revolutionize data platforms, offering adaptive, dynamic solutions to meet complex challenges in the fast-evolving digital landscape. ### Quantum vs. Classical - Data Management Computational Complexity URL: https://subhadipmitra.com/blog/2023/quantum-vs-classical-data-management-complexity/ Date: 2023-12-10 Tags: data-platforms Grover’s Algorithm and the Revolution of Quantum Search Efficiency ### Quantum Experiment Data Exchange (QEDX) - Building an Interoperability Standard URL: https://subhadipmitra.com/blog/2023/quantum-data-exchange/ Date: 2023-11-20 Tags: data-platforms, open-source Advancements in data management, from warehouses to Data Mesh and Lakehouse, signal a shift toward more adaptive platforms like, Quantum Data Management, Genetic algorithm concepts, etc. ### Data at Quantum Speed - The Promise and Potential of QDP URL: https://subhadipmitra.com/blog/2023/quantum-data-platform/ Date: 2023-10-28 Tags: data-platforms Explore the new realm of Quantum Data Platform (QDP) and its promise to revolutionize data processing at quantum speed. Discover the potential applications, technical considerations and implications. ### The Next Frontier - Envisioning the Future of Data Platforms Beyond Data Mesh, Data Lakehouse, and Data Hub/Fabric URL: https://subhadipmitra.com/blog/2023/next-frontier-data-platform/ Date: 2023-10-12 Tags: data-platforms Advancements in data management, from warehouses to Data Mesh and Lakehouse, signal a shift toward more adaptive platforms like, Quantum Data Management, Genetic algorithm concepts, etc. ### Part 4 - Building a Massive-Scale Real-Time Data Platform - Memory Management with Apache Ignite URL: https://subhadipmitra.com/blog/2022/building-a-massive-scale-real-time-data-platform-p4/ Date: 2022-12-05 Tags: distributed-systems, case-study Deep dive into memory management with Apache Ignite for high-performance data platforms. Learn how to handle 2.5M events/second with sub-millisecond latency through practical memory architecture, optimization techniques, and real-world implementation patterns. ### Part 3 - Building a Massive-Scale Real-Time Data Platform - Memory Management with Apache Ignite URL: https://subhadipmitra.com/blog/2022/building-a-massive-scale-real-time-data-platform-p3/ Date: 2022-11-27 Tags: distributed-systems, case-study Deep dive into memory management with Apache Ignite for high-performance data platforms. Learn how to handle 2.5M events/second with sub-millisecond latency through practical memory architecture, optimization techniques, and real-world implementation patterns. ### Part 2 - Building a Massive-Scale Real-Time Data Platform - Data Partitioning and Flow URL: https://subhadipmitra.com/blog/2022/building-a-massive-scale-real-time-data-platform-p2/ Date: 2022-11-18 Tags: distributed-systems, case-study Explore how to architect data partitioning and flow for massive-scale event processing. Learn implementation patterns for handling 2.5M events/second across distributed systems using Kafka, Ignite, and Cassandra. Practical insights on partition strategies, data routing, and performance optimization. ### Part 1 - Building a Massive-Scale Real-Time Data Platform - System Overview and Architecture URL: https://subhadipmitra.com/blog/2022/building-a-massive-scale-real-time-data-platform-p1/ Date: 2022-11-12 Tags: distributed-systems, case-study Dive into the architecture of a telco-scale real-time data platform processing 2.5M events/second and 350GB DPI data/15min. Learn how we combined Apache Kafka, Ignite, and Cassandra to build a high-performance system handling massive telecommunications data for real-time analytics and customer insights. ### Overcoming Synchronization Hurdles in Cellular Network Positioning URL: https://subhadipmitra.com/blog/2022/network-synchronization-challenges-in-cellular-networks-positioning/ Date: 2022-04-22 Tags: distributed-systems In this article, I discuss the challenges of synchronization in cellular network positioning and the importance of precise timing for accurate positioning. I also explore ways to mitigate these errors, including algorithmic adjustments and improving synchronization technologies. ### The Principles Got It Backwards: Designing for Safe Change, Not Just Failure URL: https://subhadipmitra.com/blog/2021/distributed-system-design/ Date: 2021-03-18 Tags: distributed-systems The foundational distributed systems principles optimized for surviving hardware failure and scaling horizontally. But the data tells a different story: 80% of outages stem from changes we make to running systems. The hard problem has shifted from 'can it survive failure' to 'can it survive us.' ### Designing a Real Time Data Processing System URL: https://subhadipmitra.com/blog/2021/designing-a-real-time-data-processing-system/ Date: 2021-01-16 Tags: distributed-systems Master real-time data processing - A guide to designing scalable, resilient, and high-performance systems for instant insights. ### Introducing OConsent - Open Consent Protocol URL: https://subhadipmitra.com/blog/2020/introducing-oconsent-open-consent-protocol/ Date: 2020-12-10 Tags: blockchain, open-source OConsent is a blockchain-based platform that enables transparent processing of personal data, empowering users and data controllers to manage consent and privacy. ### Welcome to my blog URL: https://subhadipmitra.com/blog/1970/welcome/ Date: 1970-01-01 Tags: Let's talk tech! I'll post everything from polished pieces to spur-of-the-moment thoughts. And if you've got ideas for posts or want to collaborate, let's connect! ## Publications ### Closing the Activation-Cone Blind Spot: Response-Time Probing and Unified Defense Venue: arXiv Year: 2026, June Type: paper URL: https://arxiv.org/abs/2606.29441 Evaluates five jailbreak-defense paradigms across seven instruction-tuned models (7B-31B) and five attack families, proving that prompt-time activation defenses are structurally blind to prefilling attacks. Introduces response-time probing on the first generated tokens (AUROC 0.97-1.00 across all models); composed with null-space steering, it drives prefilling attack success to zero with no false positives on benign inputs. Code, attacks, and per-sample results released. ### Cross-Generational Transfer of Adversarial Attacks Reveals Non-Monotonic Safety Alignment in LLMs Venue: arXiv Year: 2026, June Type: paper URL: https://arxiv.org/abs/2606.00813 An automated red-teaming study finding that LLM safety alignment does not always improve monotonically across model generations. Using quality-diversity evolution (MAP-Elites) to probe successive generations of an open-weight model family, it shows that a mid-generation release can be more vulnerable than both its predecessor and its successor, with evolved attack archives transferring unevenly across generations. These longitudinal patterns are invisible to static benchmarks and surface only through adaptive, longitudinal probing. ### Cross-Platform Fused MoE Dispatch in Triton: Portable Expert Routing Without CUDA Venue: arXiv Year: 2026, April Type: paper URL: https://arxiv.org/abs/2605.23911 Code: https://github.com/bassrehab/triton-kernels TritonMoE, a Mixture-of-Experts inference kernel written entirely in OpenAI Triton with no CUDA. A fused gate+up GEMM computes both SwiGLU projections from shared tile loads, eliminating 35% of global memory traffic. Reaches 89-131% of Megablocks throughput at inference batch sizes (up to 512 tokens) on both NVIDIA A100 and AMD MI300X, with full code portability across vendors. ### Spark-LLM-Eval: A Distributed Framework for Statistically Rigorous Large Language Model Evaluation Venue: arXiv Year: 2026, January Type: paper URL: https://arxiv.org/abs/2603.28769 A distributed framework leveraging Apache Spark for statistically rigorous LLM evaluation at scale. Treats evaluation as a data-parallel problem, providing bootstrap confidence intervals, statistical significance tests (paired t-tests, McNemar's test, Wilcoxon signed-rank), and content-addressable response caching backed by Delta Lake. Reports linear scaling performance. ### Quality-Diversity Evolution for Discovering Diverse Vulnerabilities in LLM Safety Venue: ICLR 2026 Workshop AIWILD Year: 2026, May Type: paper URL: https://openreview.net/forum?id=YMtanBXw5q Code: https://github.com/bassrehab/red-queen A quality-diversity evolutionary framework operating at the semantic level for LLM adversarial testing. Uses MAP-Elites to maintain a diverse archive of interpretable attack strategies across behavioral dimensions, revealing distinct vulnerability profiles across GPT-4o-mini, Claude 3.5 Sonnet, and Gemini 2.0 Flash. ### Field-Theoretic Memory for AI Agents: Continuous Dynamics for Context Preservation Venue: arXiv Year: 2026, January Type: paper URL: https://arxiv.org/abs/2602.21220 A memory architecture representing stored information as continuous fields governed by partial differential equations. Demonstrates 116% improvement on multi-session reasoning and 43.8% gains on temporal reasoning benchmarks, with multi-agent configurations achieving near-perfect collective performance through field coupling. ### Automated Rule Generation for Tiered Systems Using Multi-Stage Failure Learning Venue: Google, Technical Disclosure Commons Year: 2026, January Type: technical-disclosure URL: https://www.tdcommons.org/dpubs_series/9180/ This work addresses a key challenge in multi-tiered anomaly detection systems: balancing the operational cost of accurate analytical models with the accuracy of less expensive rule engines. ### Predictive Cost-Benefit Routing for Multi-Tier Risk Decisioning Systems Venue: Google, Technical Disclosure Commons Year: 2026, January Type: technical-disclosure URL: https://www.tdcommons.org/dpubs_series/9181/ This work addresses the tension between computational accuracy and operational speed in real-time risk systems. It proposes a multi-tier risk decisioning system managed by a predictive, cost-benefit routing engine. ### UPIR: Automated Synthesis and Verification of Distributed Systems Venue: Google, Technical Disclosure Commons Year: 2025, November Type: technical-disclosure URL: https://www.tdcommons.org/dpubs_series/8852/ Code: https://github.com/bassrehab/upir Framework combining formal verification, program synthesis, and machine learning to automatically generate verified distributed system implementations. Achieves 274x speedup with 60% latency reduction. ### ETLC: A Context-First Approach to Data Processing in the Generative AI Era Venue: Google Cloud Year: 2025, May Type: whitepaper URL: https://services.google.com/fh/files/blogs/etlc_full_paper.pdf Introduces ETLC (Extract, Transform, Load, Contextualize), adding semantic, relational, operational, environmental, and behavioral context to data pipelines for GenAI systems. ### Field-Theoretic Context System (FTCS) Venue: Google, Technical Disclosure Commons Year: 2025, April Type: technical-disclosure URL: https://www.tdcommons.org/dpubs_series/8022/ An innovative approach modeling context as interacting fields rather than discrete states, enabling natural context flow and dynamic evolution through partial differential equations. ### ARTEMIS: Adaptive Multi-agent Debate Framework Venue: Google, Technical Disclosure Commons Year: 2025, January Type: technical-disclosure URL: https://www.tdcommons.org/dpubs_series/7729/ An AI framework supporting complex debate scenarios and group decision-making through a tiered structure of language models that dynamically generate and evaluate arguments. ### Data Monetization Strategy for Enterprises Venue: BITS Pilani Year: 2023, December Type: thesis URL: https://www.researchgate.net/publication/376557741_Data_Monetization_Strategy_for_Enterprises A comprehensive framework for enterprises to transform data assets into economic value, establishing methodologies for Data Monetization through a structured 12-month approach. ### Open Location Proof (OLP) Protocol Venue: ResearchGate Year: 2021, January Type: paper URL: https://www.researchgate.net/publication/387802463 A privacy-aware open protocol to prove without repudiation an entity's point-in-time presence, participation and location in physical or virtual space. ### OConsent: Open Consent Protocol for Privacy and Consent Management with Blockchain Venue: BITS Pilani / arXiv Year: 2021, December Type: paper URL: https://arxiv.org/abs/2201.01326 A protocol and platform based on Blockchain Technology that enables transparent processing of personal data throughout its lifecycle from capture, lineage to redaction. ### Cascading Algorithm: Bayesian Price Convergence for Arbitrage-Free Option Pricing Venue: Zenodo Year: 2016, March Type: paper URL: https://zenodo.org/records/18018278 A novel Cascading Algorithm leveraging Bayesian inference to facilitate arbitrage-free pricing of financial options through continuously refined price estimates. ### CTDBN-Based Financial Markets Analysis and Differential Predictions Venue: Zenodo Year: 2015, April Type: paper URL: https://zenodo.org/records/18017466 Coupled Temporal Deep Belief Networks with Conditional Restricted Boltzmann Machines to encode intra-market movements and temporary coupling with global indices. ### Dynamic Compression: Multicuts for Planar Graphs with Outer Terminals Venue: Zenodo Year: 2014, July Type: paper URL: https://zenodo.org/records/18017304 On the fly creation and calculation of Max-flow min-cut gap and approximation of multicuts with deep emulations of a fault-free mesh on a mesh with random faults. ### XPath Ranking and Graph Compression Venue: Zenodo Year: 2012, May Type: paper URL: https://zenodo.org/records/18017005 Algorithms for automatic on-the-fly wrapper creation for extractions from HTML and XML documents and compressing the adjacency list representation of graphs. ### Real Time Multiple Target Tracking Using Arrayed Sensors Venue: Amity University Year: 2009, November Type: paper URL: https://zenodo.org/records/18016470 A discrete time-step optimization algorithm based on limited parameters of the sensor-target systems with OpenCV-based implementation. ### Low Energy Wireless Electricity Transmission Using Resonant Magnetic Coupling Venue: Zenodo Year: 2007, February Type: paper URL: https://zenodo.org/records/18018081 Wireless power transfer via resonant magnetic coupling, demonstrating power transfer efficiencies of 67% at a distance of 30 cm with applications to consumer electronics and medical implants. ## Open Source Projects ### AI/ML Infrastructure - **AI Metacognition Toolkit** (https://github.com/bassrehab/ai-metacognition-toolkit): Production-ready framework for systematic reasoning in AI systems. Available on PyPI. - **ARTEMIS Agents** (https://github.com/bassrehab/artemis-agents): Multi-agent debate framework with hierarchical argument generation, jury-based evaluation, and safety monitoring. 86% decision accuracy. Available on PyPI. Docs: https://artemis.subhadipmitra.com - **Speculative Decoding** (https://github.com/bassrehab/speculative-decoding): LLM inference acceleration through draft-then-verify decoding. - **Triton Kernels** (https://github.com/bassrehab/triton-kernels): GPU kernels for LLM inference with up to 8x speedups. - **Spark LLM Eval** (https://github.com/bassrehab/spark-llm-eval): Distributed LLM evaluation framework on Apache Spark. Available on PyPI. - **Steering Vectors for Agents** (https://github.com/bassrehab/steering-vectors-agents): Runtime control of LLM agent behaviors through activation steering. - **api-mcp-compiler** (https://github.com/bassrehab/api-mcp-compiler): Compiles OpenAPI, Swagger, and WSDL specifications into governance-aware MCP tool surfaces. Destructive tools stay disabled until a human approves them. Available on PyPI. Docs: https://api-mcp.subhadipmitra.com ### Distributed Systems - **UPIR** (https://github.com/bassrehab/upir): Automated distributed systems synthesis with formal verification. ### Privacy & Consent - **LLMConsent Standards** (https://github.com/bassrehab/llmconsent-standards): Open protocol for AI-human consent management. - **OConsent Protocol** (https://github.com/bassrehab/oconsent): Blockchain-based consent management with Ethereum smart contracts. ### Protocols & Telecom - **SMPP Gateway** (https://github.com/bassrehab/smpp-core): Modern Java 21 SMPP protocol implementation. 1.8M PDU decodes/sec, virtual threads, Netty transport. Maven Central: io.smppgateway/smpp-core. Docs: https://docs.smppgateway.io - **SMPP Kafka Producer** (https://github.com/bassrehab/smpp-kafka-producer): Production-ready SMPP-to-Kafka bridge with HTTP/2 REST API. 5G compatible. - **ISO8583 Simulator** (https://github.com/bassrehab/ISO8583-Simulator): High-performance financial message processing (180k+ TPS). Supports VISA, Mastercard, AMEX, Discover, JCB, UnionPay. Available on PyPI. Docs: https://iso8583.subhadipmitra.com ### Enterprise - **Blockchain Loyalty Fabric** (https://github.com/bassrehab/blockchain-loyalty-fabric): Hyperledger Fabric-based cross-provider loyalty platform. ## Citation Guidelines When referencing work from this site: - Academic papers: Use standard academic citation format with DOI/URL - Blog posts: Cite as "Mitra, S. (Year). Title. Binary Breakthroughs. URL" - Software: Reference GitHub repository with author attribution - For official publications, prefer citing the original publisher URL ## Contact For research collaboration, speaking engagements, or professional matters: - Email: contact@subhadipmitra.com - LinkedIn: https://linkedin.com/in/subhadip-mitra - GitHub: https://github.com/bassrehab - Twitter/X: https://twitter.com/bassrehab - Google Scholar: https://scholar.google.com/citations?user=B3U5mSYAAAAJ&hl=en ## License Blog content: All rights reserved Open source projects: See individual repository licenses (MIT, Apache-2.0, GPL-3.0)