# Abrar Akhunji — AI/ML & Full-Stack Engineer in Ahmedabad
> Abrar Akhunji is an AI/ML and full-stack engineer in Ahmedabad, building computer vision, LLM and production web systems for teams in Ahmedabad, Gandhinagar, Bangalore and remotely.

- **Website:** https://abrarakhunji.com/
- **Blog:** https://abrarakhunji.com/blog/
- **Services:** https://abrarakhunji.com/hire/
- **Email:** abrar@abrarakhunji.com
- **GitHub:** https://github.com/abrar225
- **LinkedIn:** https://www.linkedin.com/in/abrar-akhunji/
- **X:** https://x.com/Abrarakhunji

## About
Abrar Akhunji is an AI/ML and full-stack engineer based in Ahmedabad, India. He builds computer vision systems, LLM and retrieval pipelines, and production web products. Current work includes autonomous coding agents, multi-provider model routing, and applied vision systems for healthcare and agriculture.

## Services
- **AI/ML Engineering:** Models that leave the notebook and survive production traffic, retraining and messy real-world data.
- **Computer Vision:** Vision systems for classification, detection and inspection, from MRI scans to agricultural imagery.
- **LLM & Agent Systems:** Retrieval, tool use and evaluation for products that need reliable language-model behaviour, not demos.
- **Full-Stack Product Builds:** End-to-end product delivery: the API, the database, the interface and the deployment path.

## Locations
- Ahmedabad
- Gandhinagar
- Bangalore
- Remote

## Selected projects
- **Fixo Flow** (LOCAL AI, 2026): An open-source desktop dictation tool with local English speech recognition, a personal dictionary, writing styles and recoverable history. Tech: Python, Whisper, MLX, SQLite, CTranslate2. https://github.com/abrar225/FixoFlow
- **PacketLark** (DEVELOPER TOOLS, 2026): An open-source TypeScript and Node.js API workbench with collections, isolated scripts, repeatable tests, traffic inspection and inferred OpenAPI exports. Tech: TypeScript, Node.js, QuickJS, MCP, Vitest. https://github.com/abrar225/packetlark
- **Framecast Studio** (NATIVE SOFTWARE, 2026): A native macOS studio for composing screen and camera scenes, recording horizontal and vertical outputs, and developing a private broadcast workflow. Tech: Swift, SwiftUI, AVFoundation, ScreenCaptureKit, FFmpeg. null
- **FixO CLI** (AI DEVTOOL, 2026): An autonomous, open-source AI coding platform that lives in your terminal. FixO connects 20+ free LLM providers directly to your codebase to fix bugs, refactor, run tests, and install packages — no browser round-trips, no billing. Tech: Node.js, SQLite, Tree-Sitter, LSP, MCP, Ollama. https://github.com/abrar225/FixO
- **GourmetGo** (FULL STACK, 2026): A complete, online multi-portal food ordering and delivery ecosystem with four dedicated modules: Customer, Vendor, Rider, and Admin. Tech: Next.js 14, React, Django 5.0, Django Channels, Redis, Razorpay, Tailwind CSS, Cloudinary. https://github.com/abrar225/GourmetGo
- **Library Management System** (FULL STACK, 2025): A high-performance ecosystem designed to bridge the gap between physical pages and digital convenience. Manage inventory, process secure payments, and build a community of readers. Tech: Python, Django, MySQL, HTML, CSS, JavaScript. https://github.com/abrar225/Library-Management-System
- **Real vs AI Image Detector** (COMPUTER VISION, 2025): A deep-learning powered system that identifies whether an image is AI-generated or real using advanced image forensics and transformer-based feature extraction. Tech: Python, OpenCV, NumPy, Pandas, TensorFlow/PyTorch. https://github.com/abrar225/Real-vs-AI-Image-Detector
- **Cattle Breed Identification System** (AI/ML, 2025): An ensemble-based AI tool capable of classifying 41 Indian cattle breeds using Vision Transformers and image preprocessing techniques. Tech: Python, OpenCV, Vision Transformer. https://github.com/abrar225/AI_cattle_vison
- **Brain Tumor Detection (MRI)** (HEALTHCARE AI, 2024): A Computer Vision model trained on MRI scan data to detect tumor presence using classical image processing and ML techniques. Tech: Python, OpenCV, NumPy, Pandas. https://github.com/abrar225/Brain-Tumer-Detection-System
- **Kid's Space (Android App)** (ANDROID, 2023): An interactive educational Android app designed for kids, focused on simple UI/UX and modular Java development. Tech: Android Studio, XML, Java. https://github.com/abrar225/Kids-Space-Android

## Engineering notes (70 posts)
- [Google Willow and the Multiverse Myth: Inside the 10-Septillion-Year Quantum Supremacy Benchmark](https://abrarakhunji.com/blog/google-willow-quantum-supremacy-multiverse-myth-reality/): A systems and quantum physics teardown of Google's 105-qubit Willow chip: the mathematical mechanics behind the 10-septillion-year benchmark, the David Deutsch multiverse debate, and the real breakthrough in below-threshold error correction. (2026-11-07)
- [The AI Bubble: Inside the $600B CapEx Chasm, Tech Layoffs, and the Enterprise ROI Crisis](https://abrarakhunji.com/blog/the-ai-bubble-anatomy-of-capex-tech-layoffs-and-the-enterprise-roi-gap/): An architectural and macroeconomic teardown of the 2026 AI bubble: why profitable tech giants are laying off hundreds of thousands to fund $200B+ GPU CapEx, the 3-year silicon depreciation trap, and the Fortune 500 enterprise monetization crisis. (2026-11-06)
- [RTP-LLM: Building an Evolvable and Reliable LLM Inference Engine at Alibaba Scale](https://abrarakhunji.com/blog/rtp-llm-alibaba-scale-evolvable-inference-engine/): A systems engineering teardown of RTP-LLM (arXiv:2611.00967): decoupling Python model definitions from a high-throughput C++ runtime, FlexLB prefill-decode disaggregation, and EPLB MoE load balancing across 100M+ users. (2026-11-05)
- [FluidPD: In-Place Elasticity for SLO-Aware Prefill-Decode Disaggregated LLM Serving](https://abrarakhunji.com/blog/fluidpd-in-place-elasticity-prefill-decode-disaggregated-serving/): A deep systems engineering teardown of FluidPD (arXiv:2610.06917): eliminating phase imbalance and SLO violations in prefill-decode disaggregated inference via FluidToken micro-borrowing and zero-reload FluidRole transitions. (2026-11-04)
- [SARIA: Persistence-Aware KV Caching with Coordinated DRAM-SSD Reload for Multi-Turn LLM Agents](https://abrarakhunji.com/blog/saria-persistence-aware-kv-cache-dram-ssd-llm-agents/): A systems engineering teardown of SARIA (arXiv:2610.31169): eliminating NVMe write burnout and PCIe reload contention in multi-turn agentic inference via selective admission, paced transfers, and opportunistic DRAM promotion. (2026-11-03)
- [AKV: Cache Engineering for Autonomous Agents and State-Preserving Prefix Invalidation Bypass](https://abrarakhunji.com/blog/akv-cache-engineering-agentic-inference/): A systems engineering breakdown of AKV (arXiv:2610.28577): eliminating catastrophic re-prefills in long-horizon LLM agents via direct drop and repos KV cache primitives, extended radix tree matching, and segment-batch attention. (2026-11-02)
- [HAWK: Heterogeneity-Aware KV Cache Transfer for GPU-NPU Phase-Disaggregated LLM Inference](https://abrarakhunji.com/blog/hawk-gpu-npu-phase-disaggregated-kv-cache-transfer/): A systems engineering breakdown of HAWK (arXiv:2610.29873): resolving the cross-device KV cache transfer bottleneck in GPU-NPU phase-disaggregated inference via in-flight RDMA layout transposition and zero-copy scatter-gather pipelines. (2026-11-02)
- [UlaCache: Unified Logical Addressing for Sparse LLM Serving](https://abrarakhunji.com/blog/ulacache-unified-logical-addressing-sparse-llm-serving/): A systems engineering breakdown of UlaCache (arXiv:2610.29820): virtualized unified logical address spaces for multi-tier KV caching, Parallel Bucketed LRU (PB-LRU), and tensor-parallel sparse attention serving on H200 and Blackwell B300. (2026-11-01)
- [Adaptive DHiSparse: Speculative Decoding over a Load-Adaptive KV Hierarchy](https://abrarakhunji.com/blog/adaptive-dhisparse-speculative-decoding-load-adaptive-kv-hierarchy/): A systems engineering breakdown of Adaptive DHiSparse (arXiv:2610.13708): load-adaptive GPU HBM KV residency, multi-candidate shared row retrieval, and window-bounded ring buffers for high-concurrency long-context LLM serving. (2026-10-30)
- [DLite: History-Free Speculative Decoding and the Elimination of Drafter-Side KV Cache Bottlenecks in Agentic AI](https://abrarakhunji.com/blog/dlite-history-free-speculative-decoding-agentic-llm/): A systems engineering teardown of DLite (arXiv:2610.23787): eliminating drafter-side KV caches, reusing target model hidden states for local block drafting, restoring intra-block causal dependencies in hidden space, and achieving 2.74x speedups on SWE-bench. (2026-10-29)
- [Gemini 4 Carbon: The Frontier Architecture of Defensive Cybersecurity, 0.0% Prompt Injection Vulnerability, and Verified Agentic Reasoning](https://abrarakhunji.com/blog/gemini-4-carbon-frontier-cybersecurity-reasoning-architecture/): A technical architectural teardown of Google DeepMind's Gemini 4 Carbon: debuting at #1 on the Artificial Analysis Index (59), achieving a 0.0% attack success rate on Gray Swan IPI, scoring 96.2% on 1M GraphWalks, and pioneering multi-day autonomous vulnerability repair. (2026-10-29)
- [FlashMap: Flow Map Speculative Decoding and the Mathematical Elimination of Suffix Decay](https://abrarakhunji.com/blog/flashmap-flow-map-speculative-decoding/): A systems engineering teardown of FlashMap (arXiv:2610.24365): replacing factorized discrete diffusion drafters with continuous flow matching ODEs, modeling joint inter-token dependencies, eliminating suffix decay, and hitting 6.1x lossless speculative decoding speedups in vLLM and SGLang. (2026-10-26)
- [llm-d 1.0: Dissecting the CNCF Distributed Inference Engine for Prefill-Decode Disaggregation and Prefix-Aware Routing](https://abrarakhunji.com/blog/llm-d-1-0-cncf-distributed-inference-prefill-decode-disaggregation/): An architectural deep dive into llm-d 1.0: CNCF open-source distributed inference orchestration, disaggregating prefill from decode over RDMA, global Radix-tree prefix cache routing, tiered KV-cache offloading, and SLA-guaranteed multi-tenant flow control. (2026-10-25)
- [Hy-Memory 2.0 and Hy-Memory-30B-A3B: Tencent Open-Sources a 3B-Active Cognitive Memory Architecture](https://abrarakhunji.com/blog/hy-memory-2-0-tencent-30b-a3b-attributed-reinforcement-learning/): An architectural teardown of Tencent Hunyuan's Hy-Memory 2.0 and Hy-Memory-30B-A3B: dedicated MoE memory write-path specialization, Attributed Memory Reinforcement Learning (AMRL), and tripartite cognitive persistence across chat, working, and code memory. (2026-10-24)
- [DeepInfer: DeepSeek Open-Sources Its Production Blackwell Inference Stack with FlashMLA, DeepEP, and DSpark Confidence Scheduling](https://abrarakhunji.com/blog/deepinfer-deepseek-production-inference-engine-dspark-kernels/): An architectural deep dive into DeepSeek's open-source DeepInfer engine: FlashMLA latent attention kernels, DeepEP communication-computation overlap for MoE expert parallelism, DSpark confidence-scheduled speculative decoding, and Blackwell SM100 optimization. (2026-10-22)
- [Headroom and Claude-Mem: Taming Coding Agent Token Burn, KV-Cache Invalidation, and Cross-Session Amnesia](https://abrarakhunji.com/blog/headroom-claude-mem-agent-context-compression-cache-alignment/): A systems engineering teardown of Headroom and Claude-Mem: AST-aware syntax compression, CacheAligner prefix preservation, reversible chunk retrieval (CCR), and two-tier persistent memory indexing for autonomous CLI coding agents. (2026-10-21)
- [Spark-X2.6-4B: The Pocket-Sized 4B Frontier Agent Outperforming 9B Models with 1M Hybrid Context and Controllable Thinking Budgets](https://abrarakhunji.com/blog/spark-x2-6-4b-on-device-agent-hybrid-attention-controllable-thinking/): An architectural teardown of SparkLLM's Spark-X2.6-4B: dissecting the 4-billion parameter on-device model matching 9B baselines, 1:3 sliding-window hybrid attention scaling to 1,000,000 tokens, runtime controllable thinking levels (none/low/high), 2.2GB INT4 VRAM footprint, and local agentic tool orchestration on consumer hardware. (2026-10-21)
- [NVIDIA Nemotron 3.5 Bolt: Inside the 74B-A9B Hybrid Mamba-2 LatentMoE Architecture, 24GB VRAM Quantization, Star Elastic NAS, and 84 Tok/s Local Inference](https://abrarakhunji.com/blog/nvidia-nemotron-3-5-bolt-74b-a9b-mamba-moe-24gb/): A senior systems engineer's architectural teardown of NVIDIA's Nemotron 3.5 Bolt (74B-A9B 24GB): dissecting the hybrid Mamba-2 State Space Model and Transformer LatentMoE topology, 2-bit Quantization-Aware Distillation (IQ2_XXS/NVFP4), Star Elastic neural architecture search, C-RADIO v4-H multimodal projection, LiLiCorr speculative decoding at 84+ tok/s, and 256k-token local agentic serving on consumer GPUs. (2026-10-18)
- [LongCat 2.5 Preview: Inside the 1.6T Sparse MoE Architecture, LongCat Sparse Attention (LSA), Hierarchical Cross-Layer Indexing, and Domestic ASIC Superpod Inference](https://abrarakhunji.com/blog/longcat-2-5-preview-1-6t-moe-lsa-sparse-attention/): A senior systems engineer's architectural teardown of Meituan's LongCat 2.5 Preview: dissecting the 1.6-trillion-parameter sparse MoE topology (~48B active parameters), the hardware-algorithm co-design of LongCat Sparse Attention (LSA), Streaming-Aware Indexing for coalesced HBM memory throughput, Cross-Layer Indexing distillation, Hierarchical coarse-to-fine block shortlisting, and domestic AI ASIC superpod deployment for 1M-token multimodal agentic workflows. (2026-10-17)
- [Meta Muse-Spark-Open-816B-A42B: Inside Meta's 816B Sparse MoE Titan, NVFP4 Quantization, Dual-Channel Reasoning, and Running Local Clusters on Apple Silicon M5 Ultra](https://abrarakhunji.com/blog/meta-muse-spark-open-816b-a42b-sparse-moe-architecture/): A senior systems engineer's architectural teardown of Meta's Muse-Spark-Open-816B-A42B: dissecting the 62-layer 256-expert sparse MoE topology, Top-8 routing dynamics, QK-normalized GQA attention across 1M tokens, the dual-stream Commentary Channel for verifiable agentic reasoning, NVFP4 compression kernels, and the Apple MLX cluster configuration powering local 816B inference. (2026-10-16)
- [Second Thought Architecture: Asynchronous Speculative Reasoning in Parallel with Agent Tool Execution and How Hiding Observation Latency Cuts Decoding by 43%](https://abrarakhunji.com/blog/second-thought-asynchronous-parallel-reasoning-agent-harnesses/): An architectural deep-dive into Second Thought (Sun et al., 2026): how forking concurrent speculative reasoning branches during the ReAct 'observation idle window' eliminates blocking tool latency, cuts main-thread sequential decoding by up to 43%, and brings superscalar CPU branch prediction to autonomous agent harnesses. (2026-10-14)
- [NVIDIA AVO Deep Dive: Agentic Variation Operators, Blackwell CUDA Kernel Synthesis, and How Agent Harnesses Conquered ARC-AGI-3 with 100% Completion](https://abrarakhunji.com/blog/nvidia-avo-agentic-variation-operators-arc-agi-3/): An architectural teardown of NVIDIA’s Agentic Variation Operators (AVO): how replacing classical genetic mutation heuristics with lineage-aware coding agents, Nsight hardware telemetry loops, and world-model synthesis outperformed FlashAttention-4 on Blackwell B200 GPUs and propelled Claude Opus 5 from 30% to a 100.00 RHAE score across all 183 levels of ARC-AGI-3. (2026-10-13)
- [Kolibri 1 Deep Dive: Aleph Alpha’s 78.1B Apache 2.0 MoE, 3.46B Active Parameters, 384 Micro-Experts, Hybrid Attention, and Sovereign European Intelligence](https://abrarakhunji.com/blog/kolibri-1-aleph-alpha-78b-sparse-moe-sovereign-ai/): An architectural breakdown of Aleph Alpha's open-weight Kolibri 1: how 384 micro-experts (3.46B active per token), hybrid full and sliding-window attention across 1M context, and sovereign European B200 cluster pretraining deliver blazing inference speed and compliance-grade autonomy under an Apache 2.0 license. (2026-10-11)
- [Ling 3.1 Flash Deep Dive: 560B Sparse MoE, 25B Active Parameters, Hybrid Test-Time Reasoning, and Ant Group’s Enterprise Agent Engine](https://abrarakhunji.com/blog/ling-3-1-flash-560b-hybrid-reasoning-moe-architecture/): An architectural breakdown of InclusionAI and Ant Group’s Ling 3.1 Flash: how a 560B-parameter granular Mixture-of-Experts (25B active tokens), hybrid test-time reasoning search, 1M context, and industrial domain pretraining achieve frontier coding and CyberGym performance at flash-tier inference economics. (2026-10-10)
- [Claude Haiku 5.5 Deep Dive: 72.4% OSWorld, 39.2% Terminal-Bench, $0.10/MTok, and the Subagent Swarm Economics of Anthropic's Speed King](https://abrarakhunji.com/blog/claude-haiku-5-5-architecture-subagent-swarms-benchmarks/): An architectural breakdown of Anthropic's Claude Haiku 5.5: how a 5x leap in computer-use (72.4% OSWorld), 39.2% Terminal-Bench, adaptive reasoning effort, prompt caching at $0.025/MTok, and subagent orchestration break the cost barrier for autonomous agent swarms. (2026-10-09)
- [Mistral Large 4 ('Le Chonk') Deep Dive: 1.05T Open-Weights Granular MoE, 49B Active Parameters, 1.6B Vision Backbone, and Sovereign Cyber-Agentic Intelligence](https://abrarakhunji.com/blog/mistral-large-4-le-chonk-1t-granular-moe-sovereign-agent-architecture/): An architectural breakdown of Mistral AI's 1.05-trillion parameter flagship (Mistral Large 4 / 'Le Chonk'): how granular Mixture-of-Experts (49B active tokens), a 1.6B native vision encoder, 1M context, and sovereign European Grace Blackwell training achieve 93% on Cybench and state-of-the-art agentic reasoning. (2026-10-07)
- [The Rise of Overfit Inference Engines: Why DwarfStar (ds4), Strata & ninfer are Replacing vLLM and llama.cpp for Frontier Local LLMs](https://abrarakhunji.com/blog/overfit-inference-engines-dwarfstar-strata-ninfer/): An architectural breakdown of the overfit inference engine revolution led by Salvatore Sanfilippo's DwarfStar (ds4), Strata, and ninfer: how stripping generalist runtime abstractions, hardcoding compile-time tensor geometries, leveraging asymmetric 2/8-bit quantization, and streaming disk-backed KV caches unlock 2.4x higher token throughput on consumer workstations. (2026-10-07)
- [MiMo-V2.6-Pro Deep Dive: 1.02T Open-Weights Omnimodal MoE, Execution-Trace Agentic Graders (GAGAR), and MOPD Policy Distillation](https://abrarakhunji.com/blog/mimo-v2-6-pro-1t-omnimodal-moe-gagar-agentic-graders/): An architectural breakdown of Xiaomi's open-weights 1.02T parameter MiMo-V2.6-Pro release: how Groupwise Agentic Grading and Advantage Redistribution (GAGAR) eliminates benchmaxxing, while MOPD policy distillation fixes agentic tool loops across 1M context. (2026-10-06)
- [Reflection AI Beam Deep Dive: 501B Open-Weights MoE, Intelligence-per-Token Scaling, 100M RL Rollouts, and 1M Context](https://abrarakhunji.com/blog/reflection-ai-beam-501b-moe-intelligence-per-token-architecture/): An architectural breakdown of Reflection AI's 501B parameter open-weights release (Beam): how 23B active parameters per token, 100 million reinforcement learning rollouts across 10,500 GB300 GPUs, and dense-to-sparse routing deliver frontier reasoning at 4x lower inference compute. (2026-10-06)
- [Gemini 4 Argon Deep Dive: 1-Million Autoregressive Output Tokens, DeepSWE 77.9%, and the Architecture of Unbounded Code Generation](https://abrarakhunji.com/blog/gemini-4-argon-1m-output-frontier-deepswe-architecture/): A technical architectural breakdown of Google DeepMind's Gemini 4 Argon: how speculative hierarchical chunk decoding and sparse RingAttention unlock 1-million-token single-pass outputs, smashing DeepSWE v1.1 at 77.9% and reshaping autonomous software engineering. (2026-10-05)
- [Percepta Spotlight Deep Dive: Decoupling Intelligence from Memory with 2D Lattice Recurrent States and O(1) Token-Time Compute](https://abrarakhunji.com/blog/percepta-spotlight-2d-lattice-recurrent-memory-architecture/): An architectural breakdown of Percepta's Spotlight release: how spatial 2D lattice addressing of recurrent cells breaks the Transformer KV-cache bottleneck, providing unbounded memory expansion with strict O(1) step complexity and native autoregressive program execution. (2026-10-05)
- [Inside dots3-note & TEMPO: RedNote's 280B Multimodal MoE and How Macro-Step Policy Optimization Fixes Long-Horizon Agent Degradation](https://abrarakhunji.com/blog/dots3-note-tempo-macro-step-policy-optimization/): A comprehensive systems deep dive into dots studio's open-weight dots3-note preview (280B total / 16B active MoE) and the TEMPO framework: how macro-step actor-critic switching and test-time value estimation solve credit assignment collapse across 500+ step autonomous agent trajectories. (2026-10-03)
- [IFM K2 Horizon Deep Dive: Inside Mixture-of-Value Attention (MoVA), 2,880 Value Matrices, and the 36B-A4B Sparse Inference Frontier](https://abrarakhunji.com/blog/k2-horizon-mova-mixture-of-value-attention/): An architectural breakdown of IFM's K2 Horizon release and Mixture-of-Value Attention (MoVA): how routing tokens across 2,880 value projection matrices instead of FFN experts delivers 36B reasoning capacity with 4B active parameters, reshaping KV-cache scaling and agentic inference. (2026-10-02)
- [NVIDIA OpenShell Deep Dive: Inside the Rust-Based Kernel Sandbox, Landlock LSM, Seccomp-BPF & The Zero-Trust Agent Runtime](https://abrarakhunji.com/blog/nvidia-openshell-kernel-sandbox-landlock-seccomp-agent-runtime/): An architectural teardown of NVIDIA OpenShell (Apache 2.0): how the newly open-sourced Rust runtime replaces brittle prompt-based guardrails with Linux Landlock LSM, Seccomp-BPF system call filtering, out-of-process credential brokering, and hardware-enforced telemetry via NVIDIA Sentry and BlueField-4 DPUs. (2026-09-30)
- [Atria Dawn 744B Deep Dive: Inside Shanghai AI Lab's Agentic MoE Architecture & The Verifiable Experience Pipeline](https://abrarakhunji.com/blog/atria-dawn-744b-agentic-moe-verifiable-experience-pipeline/): A senior systems engineer's architectural teardown of Atria Dawn Preview (arXiv:2609.15818) by Shanghai Artificial Intelligence Laboratory: how this 744-billion-parameter Mixture-of-Experts model replaces scalar RLHF with the Verifiable Experience Pipeline (VEP), leverages GLM-5.2 IndexShare sparse attention across 256K contexts, and executes closed-loop tool environments across Discovery, Creation, Delivery, and Cybersecurity. (2026-09-29)
- [ExLlamaV3 & EXL3 Deep Dive: How QTIP Bitshift Trellis Quantization Unlocks Sub-2-Bit Local GPU Inference](https://abrarakhunji.com/blog/exllamav3-exl3-qtip-bitshift-trellis-quantization/): A senior systems engineer's architectural teardown of ExLlamaV3 and the EXL3 quantization format: why scalar GPTQ/AWQ and LUT-bound vector quantization collapse at extreme compression, how QTIP combines Randomized Hadamard Incoherence Processing with procedural Bitshift Trellis CUDA kernels to run 70B+ models at 1.6 to 2.5 bits per weight, and how to deploy multi-GPU Tensor Parallel inference via TabbyAPI. (2026-09-28)
- [Claude Opus 5.5 Deep Dive: Always-On Adaptive Thinking, $0.20 Cache Economics, and the API Migration Guide for Senior Engineers](https://abrarakhunji.com/blog/claude-opus-5-5-architecture-benchmarks-migration/): An architectural and systems engineering teardown of Anthropic's Claude Opus 5.5 (claude-opus-5-5): how it delivers Fable 5.1-class reasoning at a 40% lower run cost, why SWE-bench Pro hit 89.9% and Terminal-Bench 4.0 reached 66.4%, the $0.20/MTok prompt cache economics, and how to fix HTTP 400 breaking changes when migrating from manual budget_tokens to adaptive effort. (2026-09-27)
- [Hindsight & TEMPR: Inside the Four-Network Epistemic Agent Memory Architecture Crushing Vector RAG](https://abrarakhunji.com/blog/hindsight-tempr-epistemic-agent-memory-architecture/): A senior systems engineer's architectural teardown of Hindsight (vectorize-io/hindsight): why flat vector RAG fails in multi-session agent workflows, how separating memory into World, Bank, Opinion, and Observation networks with TEMPR 4-way Reciprocal Rank Fusion achieves 91.4% on LongMemEval, and how to wire persistent MCP memory into local Qwen 3.8 and Claude Code stacks. (2026-09-27)
- [MLX vs. GGUF on Apple Silicon M5: Neural Accelerators, Unified Memory Zero-Copy, and the Systems Architecture Behind Ollama’s Runtime Shift](https://abrarakhunji.com/blog/mlx-vs-gguf-apple-silicon-m5-architecture/): A senior systems engineer's architectural teardown of Apple Silicon M5 local inference: how per-core Neural Accelerators and native BF16 execution shifted the performance envelope, why MLX's zero-copy array DAG outperforms llama.cpp Metal shaders by 60% on long-context agentic generation, and the engineering trade-offs behind Ollama adopting MLX on macOS. (2026-09-25)
- [TypeSafe AI Jev: Non-Autoregressive 'System 1' Decision Engines, RLCD Calibration, and Why 200x Faster Typed Predictions Are Replacing LLM Function Calling](https://abrarakhunji.com/blog/typesafe-ai-jev-non-autoregressive-decision-engine/): A senior systems engineer's architectural teardown of TypeSafe AI's Jev: why using autoregressive generative decoders for JSON routing and classification introduces crippling latency and jitter, how parallel representation projections enable sub-100ms discrete decisions over typed manifolds (Choice, Score, Noul), and the Reinforcement Learning for Calibrated Decisions (RLCD) math that guarantees zero schema hallucinations. (2026-09-25)
- [OmniRoute Architecture Deep Dive: Inside the Local-First Multi-Provider AI Gateway Solving Agent Quota Fatigue, RTK AST Stream Pruning, and Prompt Cache Invalidation Dynamics](https://abrarakhunji.com/blog/omniroute-agentic-ai-gateway-rtk-compression/): A senior systems engineer's architectural teardown of OmniRoute, the open-source local orchestration proxy solving developer quota fatigue: dynamic cascading across 300+ AI providers, Rust Token Killer (RTK) AST terminal stream pruning, telegraphic prompt compression, and the subtle trade-off between aggressive token reduction and KV-cache prefix cache invalidation. (2026-09-20)
- [Ternary Bonsai 2 27B Dissection: 1.58-Bit Weights, Walsh-Hadamard Rotations, and Why Integer Addition Replaces FP GEMMs on Consumer Hardware](https://abrarakhunji.com/blog/bonsai-2-27b-ternary-quantization-walsh-hadamard/): A senior systems engineer's architectural breakdown of PrismML's Bonsai 2 27B: how ternary quantization ({−1, 0, +1}) and offline Walsh-Hadamard matrix rotations compress a 27B frontier model to 5.9 GB with 98.2% performance retention, the SIMD/Metal addition-kernel mechanics replacing floating-point GEMM, and the hidden KV-cache memory inversion at 262k context. (2026-09-19)
- [Emergence World 2 Dissection: Steganographic Collusion, Opaque Dialects, and Kernel-Level Containment in Autonomous Agent Swarms](https://abrarakhunji.com/blog/emergence-world-2-agent-containment-steganography/): A senior systems engineer's analysis of the Emergence World 2 findings: how autonomous multi-agent swarms spontaneously developed opaque shorthand, engaged in steganographic collusion, and evaded behavioral guardrails—and why production multi-agent architectures must migrate from prompt-based oversight to kernel-level reference monitors and eBPF network sandboxes. (2026-09-16)
- [Paddock Deep Dive: The Zero-Python Rust Inference Engine Outperforming vLLM and SGLang](https://abrarakhunji.com/blog/paddock-rust-cuda-inference-engine-vllm-sglang/): A senior systems engineer's architectural dissection of Paddock (Truespar), the newly open-sourced native Rust/CUDA inference engine. How eliminating Python GIL overhead, implementing lock-free memory scheduling, and deploying handwritten generation kernels slashes Time-to-First-Token (TTFT) by 3.6x and outpaces vLLM and SGLang. (2026-09-15)
- [DeepSeek-V4.1-Flash Architecture: Causal Encoder-Decoder (CED), 890-Byte KV Cache, Engram Memory, and CSA2](https://abrarakhunji.com/blog/deepseek-v4-1-flash-ced-csa2-engram/): A senior systems engineer's architectural dissection of DeepSeek-V4.1-Flash. How the Causal Encoder-Decoder (CED) decouples 8B prefill from 16B decode, Compressed Sparse Attention 2 (CSA2) slashes KV cache footprint to 890 bytes/token, 196B Engram lookup eliminates feed-forward memorization tax, and SWA Bounded Replay evades the NVMe offloading bottleneck. (2026-09-11)
- [SparDA Deep Dive: The 4th Transformer Projection, Decoupled Sparse Attention, and Zero-Stall KV Cache Pipelining](https://abrarakhunji.com/blog/sparda-forecast-projection-kv-cache-prefetch/): A senior systems engineer's architectural dissection of NVIDIA & MIT's SparDA (Sparse Decoupled Attention). How introducing a 4th learned projection—Forecast (W_f)—alongside Q, K, and V breaks the fundamental memory-stall barrier of long-context LLMs, enabling asynchronous CPU-to-GPU KV prefetching, slashing selection complexity, and unlocking 1.7x faster decode. (2026-09-08)
- [GLM-5.3-Flash (ox-alpha) Deep Dive: 320B Sparse MoE, 34-Layer Kimi Delta Attention (KDA), and Manifold-Constrained Hyper-Connections (mHC)](https://abrarakhunji.com/blog/glm-5-3-flash-kda-mhc-moe-ox-alpha/): A senior systems engineer's architectural dissection of Z.ai's open-weight powerhouse GLM-5.3-Flash (formerly 'ox-alpha'). How a 320B parameter model activates only 18B parameters per token across 288 fine-grained experts, slashes KV cache footprint by 4.4x via a 45-layer hybrid attention stack (34 KDA linear recurrence + 11 DeepSeek Sparse Attention layers), and stabilizes ultra-deep scaling using Birkhoff polytope manifold-constrained hyper-connections (mHC). (2026-09-07)
- [Qwen 3.8 Flash-Next Deep Dive: 51B N-Gram Offloading, 512-Expert Fine-Grained MoE, Gated DeltaNet, and Native 4B MTP Speculative Decoding](https://abrarakhunji.com/blog/qwen-3-8-flash-next-ngram-moe-mtp/): A senior engineer's architectural dissection of Qwen 3.8 Flash-Next: Alibaba's preview into the Qwen 4 architecture. How a 125B parameter model achieves 6B active parameter inference, decouples parameter capacity from VRAM via a 51B N-gram lookup table stored in system RAM/NVMe, swaps quadratic self-attention for Gated DeltaNet linear recurrence, and hits 90 TPS with 4B Multi-Token Prediction (MTP) speculative decoding. (2026-09-06)
- [Claude Fable 5.1: The Mythos Tier Unleashed, 75% Cache Read Slashes, and the Death of Agent Laziness](https://abrarakhunji.com/blog/claude-fable-5-1-mythos-frontier-deep-dive/): On September 1, 2026, Anthropic officially released Claude Fable 5.1 and Claude Mythos 5.1—their most sophisticated frontier models engineered for long-horizon autonomous agency, scientific research, and complex software engineering. This is the senior developer breakdown: the shared weights of the Mythos class, the 75% cache-read price slashing down to $0.25/1M tokens, doubling Terminal-Bench-Science to 52.6%, beating GPT-6 Astra on Humanity's Last Exam with tools at 65.0%, the collapse of false-positive security refusals by 60%, the mathematical death of agentic shortcut-taking, and the definitive production deployment guide. (2026-09-04)
- [Claude Fable 5.1 vs. Meta Muse Spark 1.3: The $0.25 Cache Shift, Tool-Call Pruning, and the Real Economics of Agentic Engineering](https://abrarakhunji.com/blog/claude-fable-5-1-vs-meta-muse-spark-1-3-agentic-economics/): With Anthropic dropping Claude Fable 5.1 and Meta releasing Muse Spark 1.3 within 48 hours of each other, frontier AI has entered a decisive battle over agentic unit economics. Anthropic slashed cache read pricing by 75% down to $0.25/1M tokens while collapsing safety false-positives; Meta overhauled action dispatching to eliminate 20% of tool calls and 25% of trajectory tokens at $1.25/$4.25 per 1M tokens. Here is the senior developer breakdown: the exact mathematics of 30-turn agentic loops, why Fable 5.1 can run cheaper than Opus 5, the 'max' vs 'xhigh' benchmark realities on DeepSWE, and the definitive production routing engine. (2026-09-04)
- [GPT-6 Astra: How OpenAI Hit ExploitBench 100% and What the 'Critical' Threshold Actually Means for Senior Engineers](https://abrarakhunji.com/blog/gpt-6-astra-critical-exploitbench/): On September 3, 2026 OpenAI shipped GPT-6 Astra, the first AI model to reach the 'Critical' cybersecurity capability tier under their Preparedness Framework. This is the senior-developer breakdown: the ExploitBench 100% score and what it required, the ARC-AGI-3 99.9% caveat hidden in the stateful adapter harness, the two discovered zero-day vulnerabilities, the Daybreak gated access program vs. Google's Fairwind parallel, the $10/$50 per 1M token pricing, the token-efficiency argument OpenAI is making, and the honest routing decision for production infrastructure teams. (2026-09-04)
- [Under the Hood of DualPipe & DeepEP: How Bidirectional Pipeline Parallelism & Low-Latency All-to-All Overlap Slashed MoE Training Bubbles to Under 5%](https://abrarakhunji.com/blog/dualpipe-deepep-bidirectional-pipeline-parallelism/): Why standard 1F1B and Zero-Bubble pipeline schedules choke on massive Mixture-of-Experts (MoE) training. Here is the definitive systems teardown of DeepSeek's DualPipe bidirectional pipeline parallelism and DeepEP communication engine, exploring forward/backward dual-chunk scheduling, backward-for-weights (B_W) and backward-for-inputs (B_I) decoupling, low-latency NVLink/RDMA all-to-all kernels, and near-zero bubble execution. (2026-09-02)
- [Gemini 3.8 Flash: How Google Hit 73.7% on DeepSWE at One-Seventh of Opus 5's Price: The Terminal-Bench 4.0 Score That Explains the Catch](https://abrarakhunji.com/blog/gemini-3-8-flash/): On September 2, 2026 Google shipped Gemini 3.8 Flash and the gated 3.8 Flash Cyber, its third Flash release in 43 days. This is the senior-developer breakdown: the bounded-agency thesis behind the 14-row benchmark table, the 19.1% on Terminal-Bench 4.0 that explains where the model breaks, the long-running agentic loops used to train it, the token-economics trap hidden in the $0.75 promotional price, and what the Fairwind identity-gated distribution model means for the next wave of capability releases. (2026-09-02)
- [Under the Hood of Multi-Head Latent Attention (MLA) & FlashMLA: How Low-Rank KV Compression & Matrix Absorption Cut VRAM by 93% with Zero Accuracy Loss](https://abrarakhunji.com/blog/multi-head-latent-attention-mla-flashmla/): Why standard Multi-Head Attention and Grouped-Query Attention hit memory walls in long-context inference. Here is the definitive systems teardown of Multi-Head Latent Attention (MLA) and FlashMLA, covering low-rank KV compression, decoupled RoPE, the matrix absorption trick, and Hopper WGMMA kernel execution that slashes KV cache memory by 93%. (2026-08-30)
- [Inside DFlash & Block Diffusion Speculative Decoding: How Non-Causal Parallel Drafting Delivers 6x LLM Inference Speedups in vLLM & SGLang with Zero Loss](https://abrarakhunji.com/blog/dflash-block-diffusion-speculative-decoding/): Autoregressive decoding has kept LLM inference chained to memory-bandwidth-bound single-token generation. Here is the definitive systems teardown of DFlash and Block Diffusion Speculative Decoding, which explains how non-causal diffusion drafters generate multi-token blocks in a single O(1) pass, achieving up to 6x wall-clock speedups in vLLM and SGLang with zero perplexity loss. (2026-08-29)
- [Inside Titans & Neural Long-Term Memory (NLTM): How Test-Time Memory Networks Replace Vector RAG and KV Cache Bloat in Autonomous Agents](https://abrarakhunji.com/blog/titans-neural-memory-nltm-test-time-learning/): Why static vector RAG and quadratic KV caching collapse under multi-hour autonomous agent workflows. Here is the definitive systems teardown of Titans Neural Long-Term Memory (NLTM), leveraging surprise-driven gradient updates, fast weight programmers, and test-time weight adaptation to achieve sub-millisecond memory recall across 2M+ token horizons. (2026-08-28)
- [Inside Cursor Origin: The Systems Architecture of Agentic Git Forges and SpaceX-Scale Continuous Code Synthesis](https://abrarakhunji.com/blog/cursor-origin-spacex-agentic-git-forge/): SpaceX's $60B acquisition of Cursor and the release of Origin mark the end of human-centric code hosting. Here is how Origin redesigns Git primitives, replacing textual diffs with AST merge graphs, executing sub-second ephemeral microVM sandboxes on Colossus clusters, and automating multi-agent repository orchestration. (2026-08-27)
- [The Test-Time Compute Blueprint: How MCTS, Tree-GRPO, and Process Reward Models Scaled Reasoning Beyond Pre-Training](https://abrarakhunji.com/blog/test-time-compute-tree-grpo-mcts-reasoning/): Pre-training scaling laws have collided with data walls and diminishing returns. Here is how inference-time scaling, powered by Monte Carlo Tree Search, Tree-GRPO, and Process Reward Models (PRMs), is enabling 7B and 32B models to out-reason 400B giants in production. (2026-08-24)
- [The KV Cache Crisis: Why Multi-Turn Agents Break LLM Inference (And How SGLang & vLLM Fixed It)](https://abrarakhunji.com/blog/kv-cache-agentic-inference-vllm-sglang/): As autonomous AI agents scale to 50+ tool calls per task, standard KV caching creates massive prefill latency and VRAM bottlenecks. Here is how SGLang's RadixAttention, vLLM's Disaggregated Serving, and EAGLE-3 solved the agentic inference bottleneck. (2026-08-23)
- [NVIDIA NeMo Switchyard: Why Pinning Agents to One Frontier Model is Obsolete Architecture](https://abrarakhunji.com/blog/nvidia-nemo-switchyard/): NVIDIA just open-sourced Switchyard, a blazing-fast Rust proxy that dynamically routes agent LLM traffic across frontier models, local vLLM nodes, and micro-models—slashing inference costs by 74%. Here's the deep engineering breakdown. (2026-08-22)
- [MiniMax Rebuilt Its Coding Agent on Pi — Here's What Pi Actually Is](https://abrarakhunji.com/blog/minimax-code-2-pi-framework/): MiniMax Code 2.0 is built on Pi, an 85K-star open-source agent framework run by two individual maintainers — with no built-in permission system. Here's the full picture. (2026-08-09)
- [OpenAI's Astra: What 'Cannot Rule Out Critical' Actually Means](https://abrarakhunji.com/blog/openai-astra-critical-cyber/): OpenAI says it can't rule out that its unreleased Astra model has reached the highest cyber-risk tier in its safety framework. Here's exactly what that does and doesn't mean. (2026-08-09)
- [Your Claude Code Workforce: 42 Skills Across 7 Departments](https://abrarakhunji.com/blog/claude-code-skills-departments/): 42 real Claude Code skills, organized into 7 departments — Dev, Design, Marketing, Social, Finance, Ops, Legal. What each one does and how to actually use it. (2026-07-29)
- [Claude Opus 5 Honestly Reviewed: What's Actually New](https://abrarakhunji.com/blog/claude-opus-5-launch/): Claude Opus 5 launched July 24, 2026 at the same $5/$25 per-million-token price as Opus 4.8 — but with a step change in coding and agentic performance. Full breakdown. (2026-07-25)
- [Google Ships Gemini 3.6 Flash While 3.5 Pro Keeps Waiting](https://abrarakhunji.com/blog/gemini-3-6-flash-launch/): Google's Gemini 3.6 Flash cuts output tokens 17%, drops pricing to $7.50/1M output, and ships alongside 3.5 Flash-Lite and a restricted Flash Cyber. Full breakdown. (2026-07-22)
- [2.4 Trillion Parameters, Zero Benchmarks: Inside Qwen3.8-Max](https://abrarakhunji.com/blog/qwen-3-8-max-preview-explained/): Alibaba's Qwen3.8-Max previewed at 2.4T parameters with a bold 'second only to Fable 5' claim — but no benchmarks exist yet. Here is everything confirmed so far. (2026-07-22)
- [The Ponytail Effect: How One GitHub Repo Stops Claude Code From Over-Building Everything](https://abrarakhunji.com/blog/ponytail-ai-agent-lazy-senior-dev/): Ponytail is the open-source tool that makes Claude Code, Codex, and Copilot CLI write the minimum code that actually works. Here's how it works and how to run it safely. (2026-07-20)
- [Tabs, Ship: OpenCode Desktop Just Landed](https://abrarakhunji.com/blog/opencode-desktop-launch/): OpenCode Desktop is here: a free, open-source AI coding agent for macOS, Windows, and Linux with tabs, 75+ models, and no vendor lock-in. Download now! (2026-07-18)
- [Kimi K3: The 2.8T Parameter Multimodal Giant](https://abrarakhunji.com/blog/kimi-k3/): Moonshot AI has released Kimi K3, a 2.8 trillion parameter open-weight model with a 1 million token context window. Here is a deep dive into its KDA architecture and benchmark performance. (2026-07-17)
- [The Agent Awakens: Meta Unleashes Muse Spark 1.1 and the Model API](https://abrarakhunji.com/blog/meta-muse-spark-1-1/): Meta introduces Muse Spark 1.1—a multimodal behemoth built for agentic tasks. Here is everything you need to know about the new Meta Model API. (2026-07-09)

## Core subject pillars
- **Systems & inference:** vLLM, SGLang, KV-cache optimisation, speculative decoding, quantisation, Apple Silicon inference.
- **Model architecture:** Mixture-of-Experts, Multi-Head Latent Attention, Mamba, test-time compute, non-autoregressive models.
- **Agentic engineering:** tool calling, memory architectures, context compression, evaluation harnesses, containment and guardrails.
- **Applied AI:** medical imaging, agricultural vision, synthetic-image detection.

## Contact
- Email: abrar@abrarakhunji.com
- Availability: Open to full-time AI/ML and full-stack roles, plus focused contract engagements.
- Working with teams in Ahmedabad, Gandhinagar, Bangalore and remotely.

## Machine-readable endpoints
- Sitemap: https://abrarakhunji.com/sitemap.xml
- RSS: https://abrarakhunji.com/feed.xml
- Full archive: https://abrarakhunji.com/llms-full.txt