https://mudassiruddin7.github.io/blogs/ What I built, what broke, what I learned. en https://mudassiruddin7.github.io/blogs/three-certificates/ https://mudassiruddin7.github.io/blogs/three-certificates/ Tue, 01 Sep 2026 00:00:00 +0000 CS50x, Anthropic's Claude architect exam and two Neo4j certifications. None of them is research. Each covered a specific layer of what I build. https://mudassiruddin7.github.io/blogs/bkf-layer-order/ https://mudassiruddin7.github.io/blogs/bkf-layer-order/ Tue, 01 Sep 2026 00:00:00 +0000 Our ICLR 2027 submission: 49 small models, 15 designs, one mechanism changed at a time. Attention-sink size did not predict recall. Layer order and key binding did. https://mudassiruddin7.github.io/blogs/four-upgrades-that-dont-add-up/ https://mudassiruddin7.github.io/blogs/four-upgrades-that-dont-add-up/ Tue, 01 Sep 2026 00:00:00 +0000 LAF-YOLOv10 combined four published small-object detection techniques in one detector. Together they cost 7.8 points of mAP, and the loss traces to one interaction. https://mudassiruddin7.github.io/blogs/latex-intelligence/ https://mudassiruddin7.github.io/blogs/latex-intelligence/ Tue, 01 Sep 2026 00:00:00 +0000 Nine months of work, now on the VS Code Marketplace: editable equations, live cross-references, and failed builds explained in plain words. https://mudassiruddin7.github.io/blogs/instructions-hidden-in-log-fields/ https://mudassiruddin7.github.io/blogs/instructions-hidden-in-log-fields/ Tue, 01 Sep 2026 00:00:00 +0000 Guardian-AI v3.0 sits between untrusted security telemetry and the LLM that reads it. A regex list catches the obvious attacks. The rest need layers. https://mudassiruddin7.github.io/blogs/alphaearth/ https://mudassiruddin7.github.io/blogs/alphaearth/ Tue, 01 Sep 2026 00:00:00 +0000 AlphaEarth goes from cloud-masked composites to damage tiers for each parcel, and scores itself against independent damage records. https://mudassiruddin7.github.io/blogs/text-never-leaves-the-device/ https://mudassiruddin7.github.io/blogs/text-never-leaves-the-device/ Sat, 01 Aug 2026 00:00:00 +0000 Collecting Telugu, Tamil, Hindi and Bengali text for training without shipping anyone's health records to a server. The rule is one line; enforcing it took four separate guards. https://mudassiruddin7.github.io/blogs/i-audited-my-own-paper/ https://mudassiruddin7.github.io/blogs/i-audited-my-own-paper/ Sat, 01 Aug 2026 00:00:00 +0000 HAGD's search is fast. The circuit it found in GPT-2 Small kept no more behaviour than random features, and the preprint's larger results were not in the code. What the audit found, and what I changed. https://mudassiruddin7.github.io/blogs/what-the-agentcompress-notebook-shows/ https://mudassiruddin7.github.io/blogs/what-the-agentcompress-notebook-shows/ Wed, 01 Jul 2026 00:00:00 +0000 The paper's benchmark is not public, so the released notebook runs on synthetic workflows and its numbers are lower. Both columns belong in the README. https://mudassiruddin7.github.io/blogs/four-proposals/ https://mudassiruddin7.github.io/blogs/four-proposals/ Wed, 01 Jul 2026 00:00:00 +0000 What the proposals we sent to Google, Amazon and GARA this year each set out to answer, including the one that names its own likely negative result. https://mudassiruddin7.github.io/blogs/vak-sarvam/ https://mudassiruddin7.github.io/blogs/vak-sarvam/ Wed, 01 Jul 2026 00:00:00 +0000 An AI literacy and interpretability platform built natively in Telugu and Hindi, not translated from an English tool. https://mudassiruddin7.github.io/blogs/done-with-the-degree/ https://mudassiruddin7.github.io/blogs/done-with-the-degree/ Mon, 01 Jun 2026 00:00:00 +0000 Three years of diploma, three years of engineering, and the two habits they gave me. https://mudassiruddin7.github.io/blogs/does-the-explanation-predict-the-model/ https://mudassiruddin7.github.io/blogs/does-the-explanation-predict-the-model/ Fri, 01 May 2026 00:00:00 +0000 In May I began working with Prof. Hamdi Al-Jamimi at KFUPM on a narrow question about sepsis decision support: when a model explains itself, is the explanation true? https://mudassiruddin7.github.io/blogs/flood-claims-from-radar/ https://mudassiruddin7.github.io/blogs/flood-claims-from-radar/ Fri, 01 May 2026 00:00:00 +0000 At the 418 Hackathon at Mahindra University we built OrbitClaim Sentinel: Sentinel-1 radar in, a flood-claim decision out, in a JSON small enough to downlink. https://mudassiruddin7.github.io/blogs/below-the-framework/ https://mudassiruddin7.github.io/blogs/below-the-framework/ Wed, 01 Apr 2026 00:00:00 +0000 We were preparing proposals for the Google TPU Research Awards and the Amazon Research Awards, so I asked for collaborators who care about memory hierarchies and kernels. https://mudassiruddin7.github.io/blogs/a-cost-check-for-agent-code/ https://mudassiruddin7.github.io/blogs/a-cost-check-for-agent-code/ Wed, 01 Apr 2026 00:00:00 +0000 At the OpenAI Codex Community Hackathon in Hyderabad we built Gravity, a FinOps check that runs before AI-written code is merged. https://mudassiruddin7.github.io/blogs/overdue-is-not-a-status/ https://mudassiruddin7.github.io/blogs/overdue-is-not-a-status/ Sun, 01 Mar 2026 00:00:00 +0000 A small tool for compliance analysts, and the one modelling decision in it that I would defend in any review. https://mudassiruddin7.github.io/blogs/pygam-p-values/ https://mudassiruddin7.github.io/blogs/pygam-p-values/ Sun, 01 Mar 2026 00:00:00 +0000 Generalized additive models report p-values for their smooth terms. Getting those right needs Wood's rank-r eigendecomposition. https://mudassiruddin7.github.io/blogs/eleven-pull-requests-to-pgmpy/ https://mudassiruddin7.github.io/blogs/eleven-pull-requests-to-pgmpy/ Sun, 01 Feb 2026 00:00:00 +0000 Reading a causal-inference library closely enough to fix it, from d-separators to NaN treatment effects. https://mudassiruddin7.github.io/blogs/pihole-tailscale/ https://mudassiruddin7.github.io/blogs/pihole-tailscale/ Sun, 01 Feb 2026 00:00:00 +0000 Pi-hole filters ads at the DNS level at home, and Tailscale carries the same filter to my phone when I am out. About two hours of setup. https://mudassiruddin7.github.io/blogs/ai-chief-of-staff/ https://mudassiruddin7.github.io/blogs/ai-chief-of-staff/ Sun, 01 Feb 2026 00:00:00 +0000 At the Hack-Nation Global AI Hackathon with the MIT Sloan AI Club we mapped how information moves inside a company. The RAG pipeline died at hour 22. https://mudassiruddin7.github.io/blogs/searching-a-hierarchy/ https://mudassiruddin7.github.io/blogs/searching-a-hierarchy/ Thu, 01 Jan 2026 00:00:00 +0000 HAGD, our circuit-extraction preprint: cluster the attribution graph, search it from the top, and report a circuit only after it passes causal tests. https://mudassiruddin7.github.io/blogs/agentcompress/ https://mudassiruddin7.github.io/blogs/agentcompress/ Thu, 01 Jan 2026 00:00:00 +0000 AgentCompress picks a precision for each step of an LLM agent's workflow, from the first 32 tokens of the request. https://mudassiruddin7.github.io/blogs/pruning-for-the-disease/ https://mudassiruddin7.github.io/blogs/pruning-for-the-disease/ Thu, 01 Jan 2026 00:00:00 +0000 Our first preprint of the year keeps the channels that separate one leaf disease from another, so a few-shot model fits on a Raspberry Pi 4. https://mudassiruddin7.github.io/blogs/a-pdf-in-a-training-series-out/ https://mudassiruddin7.github.io/blogs/a-pdf-in-a-training-series-out/ Mon, 01 Dec 2025 00:00:00 +0000 At the ElevenLabs Worldwide Hackathon in Hyderabad we turned long training documents into short narrated video episodes. https://mudassiruddin7.github.io/blogs/agentathon-2025/ https://mudassiruddin7.github.io/blogs/agentathon-2025/ Mon, 01 Dec 2025 00:00:00 +0000 Agentathon 2025 set a Guinness World Record for the largest agentic AI hackathon. What I built there, and the one design idea I kept.