# Yadunandan Singh Self-taught AI/ML developer and B.Tech Artificial Intelligence & Data Science student in India. ## Profile - Name: Yadunandan Singh - Role: AI / Machine-Learning Engineer - Location: India (Asia/Kolkata (UTC+05:30)) - Email: yadunandansingh027@gmail.com - Website: https://yadunandan-singh.pages.dev - GitHub: https://github.com/YADUNANDAN-SINGH - LinkedIn: https://www.linkedin.com/in/yadunandan-singh-ai-dev - Medium: https://medium.com/@yadunandan-ai-dev Yadunandan Singh is a self-taught AI and machine-learning developer based in India, and a B.Tech student in Artificial Intelligence & Data Science at CTITR (CT Group of Institutions), Jalandhar, which he began in 2026. He works to a "no black boxes" approach: before reaching for a library, he rebuilds the core component from scratch to understand it — a vector store with hand-written cosine similarity, an SVD recommender, a Bitcoin wallet from cryptographic primitives. His public track record is 22 merged pull requests in the c2siorg organization since June 2026, across the DataLoom data-cleaning tool and the TensorMap model builder, spanning background job execution, snapshot-based undo/redo, backend performance and CI regression benchmarking, server-side pagination, CI security, pandas transformations and React UI work. Approach: "No black boxes": rebuilds core components from scratch (vector store, recommender, crypto wallet) before relying on libraries. ## Education - B.Tech in Artificial Intelligence & Data Science — CTITR (CT Group of Institutions), Jalandhar, began 2026 (in progress) - 18.06 Linear Algebra — MIT OpenCourseWare (self-directed) (completed) ## Awards - Second place, CT BuildFest 2.0 cybersecurity hackathon (CT Group of Institutions, Jalandhar, 2026-10-31): Wrote the code for Veracity, a live scam-call and AI-voice detector, over 30 hours for the six-member team Coffee to Code. The story: https://yadunandan-singh.pages.dev/blog/first-hackathon-scam-call-detector/ ## Technical skills - Languages: Python, TypeScript / JavaScript, Rust (learning; used for Prism-RAG's vector store and bitcoin-wallet-rs), SQL - Backend & data: FastAPI, pandas (profiling, transformations, time-series resampling), PostgreSQL, SQLAlchemy & Alembic, background job execution, REST API design, server-side pagination - Frontend: React, Vite, Tailwind CSS, theming with semantic CSS tokens - AI / ML: retrieval-augmented generation (RAG), vector search and cosine similarity, sentence-transformers embeddings, local LLM serving with Ollama and streaming responses, TensorFlow (CNNs), SVD-based recommender systems, linear algebra (completed MIT OCW 18.06) - Systems & tooling: Axum, Tokio and ndarray (Rust), WebAssembly, Docker Compose multi-service setups, Git (rebase, squash, fork workflows), Bitcoin primitives (secp256k1, P2WPKH, Esplora API on signet) - CI/CD & testing: GitHub Actions (including secure workflow design for fork PRs), pytest & pytest-benchmark (CI regression gates), Vitest, Playwright end-to-end tests ## Projects - [Prism-RAG](https://github.com/YADUNANDAN-SINGH/Prism-RAG): Privacy-first local RAG engine with a from-scratch Rust vector store, running entirely on your own machine. Stack: Rust (Axum, Tokio, ndarray), Python (FastAPI, sentence-transformers), React, Ollama, Docker Compose - [Veracity](https://github.com/YADUNANDAN-SINGH/Veracity): Live scam-call and AI-voice detector for ordinary phone calls, running entirely offline on a laptop, with a PayPal payment gate. Stack: Python (FastAPI, faster-whisper, PyTorch Spectra-0 / wav2vec2 XLS-R), Ollama (Llama 3.1 8B), PayPal Orders & Payments APIs, React, SQLite, scrcpy, BlackHole - [GlassBox](https://github.com/YADUNANDAN-SINGH/GlassBox-Rust-SVD-recommendation-system): Zero-knowledge recommendation engine running client-side in the browser, with SVD compiled to WebAssembly. Stack: Rust, WebAssembly, Leptos, SurrealDB - [bitcoin-wallet-rs](https://github.com/YADUNANDAN-SINGH/bitcoin-wallet-rs): Bitcoin wallet built up from cryptographic primitives rather than a wallet library. Stack: Rust, secp256k1, Esplora API - [Neuro-Symbolic Math Solver](https://github.com/YADUNANDAN-SINGH/Neuro-Symbolic-Solver): Solves handwritten equations by pairing a TensorFlow CNN recogniser with a deterministic symbolic engine, so its answers are computed, not generated. Stack: TensorFlow, FastAPI, React, Docker - [Django SVD Face Auth](https://github.com/YADUNANDAN-SINGH/django-svd-face-auth): Face-recognition login built on Singular Value Decomposition instead of a pretrained face-recognition library. Stack: Django, OpenCV, NumPy, Docker - [SVD-Based YouTube Recommender](https://github.com/YADUNANDAN-SINGH/YouTube-video-recommendation-model-with-SVD): Content-based video recommender using TF-IDF and Truncated SVD. Stack: Flask, scikit-learn, SVD - Rust Actix Quiz Game (private repository): Quiz service with server-side score validation, built on Rust and Actix Web. Stack: Rust, Actix Web, Vanilla JS, Docker - [Geometric Transformation Visualizer](https://github.com/YADUNANDAN-SINGH/Geometric-Transformation-Visualizer-): Interactive tool that visualises 2D matrix transformations in real time. Stack: Python, NumPy, Matplotlib - [SVD Image Compressor](https://github.com/YADUNANDAN-SINGH/Image_Compression_via_Low-Rank_Matrix-_Approximation): Image compression implemented from scratch through low-rank matrix approximation. Stack: Python, NumPy, Flask - [AI Image Recognition Web App](https://github.com/YADUNANDAN-SINGH/AI-Image-Recognition-webapp): CNN image classifier trained on CIFAR-10 and served through a web API. Stack: Flask, TensorFlow, JavaScript - [Delhi NCR Property Price Predictor](https://github.com/YADUNANDAN-SINGH/Delhi-NCR-property-price-predictor): Machine-learning pipeline predicting residential property prices in the Delhi NCR region. Stack: Python, pandas, scikit-learn ## Open-source contributions 22 merged pull requests in the c2siorg organization since June 2026 (21 in DataLoom, 1 in TensorMap), all merged by maintainer @ivantha. Data from GitHub, updated 14 Nov 2026. Repositories: https://github.com/c2siorg/dataloom (Open-source data-transformation platform (FastAPI + React)), https://github.com/c2siorg/tensormap (Open-source platform for building ML models without writing code). Full record with write-ups: https://yadunandan-singh.pages.dev/open-source/ - [DataLoom #563](https://github.com/c2siorg/dataloom/pull/563) — "perf(backend): add pytest-benchmark suite that fails PRs on regressions" (Performance; closes #562), merged 2026-10-28, +229/−0 in 7 files. Added an 11-benchmark `pytest-benchmark` suite on DataLoom's backend data path, plus a pull-request CI workflow that compares against `main` and fails on regressions above 50%. Result (from the PR): Baseline mins on the 100k-row CSV range from 1.32 ms (`test_infer_datetime_columns[order_id]`) to 128 ms (`test_read_csv_uncached`, with `test_assess_quality` at 65.6 ms). Reverting #517's datetime pre-gate failed the check as expected (+3,226%, +3,305% and +127%), while A/A runs on unchanged code passed (worst +7.2% locally with warmup, +16.8% across three runs on a 2-vCPU Linux VM). - [DataLoom #540](https://github.com/c2siorg/dataloom/pull/540) — "feat: run pipeline Runs and reverts as background jobs with progress and cancel" (Feature; part of #537), merged 2026-10-27, +3835/−257 in 46 files. Moved pipeline Runs and checkpoint reverts into background jobs backed by a `jobs` table and an in-process executor, with `202` submission, live step progress, cooperative cancel and held workspace reads (phase 1 of #537). Result (from the PR): Live smoke test on a 100 MB (1.1M-row) CSV with a 180-step pipeline (Apple M5, PostgreSQL 16): `main` timed out on the client at 30.0 s and committed at 36.9 s (retry → 400); the job endpoint returned `202` in 14 ms, reported 38 progress updates and succeeded at 39.6 s (revert: `202` in 6 ms). Cancelling at step 22 stopped in 0.53 s with SHA-256 and change log unchanged; SIGTERM exited cleanly in 0.3 s. - [DataLoom #554](https://github.com/c2siorg/dataloom/pull/554) — "perf(frontend): virtualize columns in Table for wide datasets" (Performance; closes #551), merged 2026-10-13, +696/−7 in 4 files. Above 50 data columns, the Table renders only the columns inside the horizontal viewport, so a wide CSV no longer puts tens of thousands of cells in the DOM. - [DataLoom #536](https://github.com/c2siorg/dataloom/pull/536) — "feat: enforce upload limit while streaming and sample-gate profiling" (Feature; part of #533), merged 2026-10-13, +1364/−70 in 30 files. Enforced the upload size limit while the request body streams in, and added sampled pre-gates that let profiling skip work a sample proves unnecessary (parts 1 and 3 of #533). Result (from the PR): On the PR's synthetic CSVs (2 vCPU / 7 GB Linux, Python 3.12.3, pandas 3.0.1): `assess_quality` went from 15.9 s to 1.7 s at 10 MB and from 155.1 s to 14.5 s at 100 MB, with byte-identical JSON output and unchanged peak memory. - [DataLoom #526](https://github.com/c2siorg/dataloom/pull/526) — "feat: snapshot-based undo and redo instead of replaying the change log" (Feature; closes #525), merged 2026-10-02, +2026/−171 in 29 files. Replaced replay-based undo with byte-exact snapshots, added Redo, and fixed the three undo bugs described in #525. Result (from the PR): Undoing the latest step on a 5 MB, 100k-row CSV took 370 ms to 960 ms on `main` (1 to 50 logged steps) and 33 ms to 43 ms with snapshots; redo takes about 37 ms. - [DataLoom #488](https://github.com/c2siorg/dataloom/pull/488) — "feat: add a standardizeDates transformation for columns with mixed date formats" (Feature; closes #487), merged 2026-09-29, +668/−2 in 11 files. Added a `standardizeDates` transformation that rewrites a mixed-format date column into one format, and wired it to the Quality panel's date-format finding, which previously had no fix button. - [DataLoom #521](https://github.com/c2siorg/dataloom/pull/521) — "perf(backend): cache parsed DataFrames keyed on path, mtime and size" (Performance; closes #520), merged 2026-09-27, +546/−4 in 9 files. Added a process-local, size-bounded LRU cache in front of `read_table_safe`, so repeat reads of an unchanged dataset file skip the parse and the datetime inference. Result (from the PR): On a 200,000-row × 5-column CSV (8.3 MB), `read_table_safe` took 282.7 ms cold and 0.091 ms on a cache hit (mean of 20). - [DataLoom #517](https://github.com/c2siorg/dataloom/pull/517) — "perf(backend): sampled pre-gate for datetime inference" (Performance; closes #515), merged 2026-09-23, +191/−3 in 2 files. `_infer_datetime_columns` ran a full regex-and-parse pass over every value of every object column on every dataset read, even for plainly non-date columns. Added a seeded 1,000-value random sample that pre-screens each column before the full inference runs, so only plausible date columns pay for it; avoids full-column scans on non-date columns without changing conversion behaviour. Result (from the PR): Best of 3 on Apple M5, Python 3.13, pandas 3.0.1: a 1M-row free-text or order-ID column went from 0.481 s to 0.007 s (69×), a 100k-row text column from 0.052 s to 0.002 s (26×), and a 1M-row column of real dates was unaffected (0.683 s vs 0.701 s, within noise). - [DataLoom #507](https://github.com/c2siorg/dataloom/pull/507) — "[Fix] Paginate every write-path endpoint instead of returning the full dataset" (Fix; closes #506), merged 2026-09-09, +828/−116 in 26 files. Found that every write-path endpoint — upload, transform-apply, save, undo, revert, file-append and pipeline-apply — returned the full dataset instead of the requested page, because the pagination logic only ran in preview mode. Lifted it into a shared `paginate_dataframe()` helper applied uniformly across the endpoints, with matching frontend changes. Result (from the PR): On a 121,500-row project, save, undo, revert, upload, file-append and pipeline-apply each returned about 9 MB, up to 2,200× more data than the client asked for; they now return only the requested page. - [DataLoom #465](https://github.com/c2siorg/dataloom/pull/465) — "feat: add profiling options dropdown to workspace tab bar '+' button" (Feature; closes #464), merged 2026-08-08, +451/−37 in 8 files. Made the workspace tab bar's '+' button open a dropdown of the profiling options, so the controls are reachable from the tab bar rather than only the ribbon menu. - [DataLoom #472](https://github.com/c2siorg/dataloom/pull/472) — "feat: add edit button () and modal to dataset cards on Homescreen" (Feature; closes #471), merged 2026-08-03, +573/−31 in 6 files. Gave the Homescreen dataset cards an overflow (kebab) menu holding Edit and Delete, plus an Edit Project modal, reducing the number of controls visible on each card. He proposed the change in issue #471 and revised the design after maintainer feedback. - [DataLoom #470](https://github.com/c2siorg/dataloom/pull/470) — "feat: download rendered charts and the correlation matrix as PNG" (Feature; closes #469), merged 2026-08-03, +788/−17 in 8 files. Added PNG export for rendered charts and for the correlation matrix, so analysis produced in the app can be taken out of it. - [DataLoom #467](https://github.com/c2siorg/dataloom/pull/467) — "feat(auth): add live password strength meter to password creation forms" (Feature; addresses #466), merged 2026-08-03, +86/−2 in 6 files. Added a reusable `PasswordStrengthMeter` component giving live feedback as a password is typed, wired into the sign-up, reset-password and change-password forms. - [DataLoom #458](https://github.com/c2siorg/dataloom/pull/458) — "feat: Add instant hover tooltips to MenuNavbar buttons and tabs" (Feature), merged 2026-07-25, +238/−21 in 5 files. Replaced the browser's delayed native tooltips on the MenuNavbar buttons and tabs with instant hover tooltips, and wrote explanations of what each control does. - [DataLoom #460](https://github.com/c2siorg/dataloom/pull/460) — "feat: add quick theme toggle button to top navbar" (Feature; closes #459), merged 2026-07-24, +15/−1 in 1 file. Added a quick light/dark theme toggle to the top navbar, replacing a multi-step trip through the settings panel. - [DataLoom #453](https://github.com/c2siorg/dataloom/pull/453) — "fix: implement proper dark mode styling for Quality tab and assessment panel" (Fix; closes #452), merged 2026-07-23, +130/−36 in 4 files. Fixed dark mode on the Quality tab and Quality Assessment panel by replacing hardcoded Tailwind light classes with semantic CSS theme tokens, so the components follow the active theme. - [DataLoom #455](https://github.com/c2siorg/dataloom/pull/455) — "fix: run lint on pull_request and comment via workflow_run" (CI; closes #454), merged 2026-07-23, +201/−106 in 2 files. A CI security fix. Split the lint workflow along the trust boundary: linting runs on `pull_request` with read-only permissions, while a separate `workflow_run` job holds `pull-requests: write` to post comments. This closed a "pwn request" privilege-escalation risk from fork pull requests instead of disabling the checkout guard. - [DataLoom #408](https://github.com/c2siorg/dataloom/pull/408) — "feat: add hover tooltips to toolbar icons" (Feature; closes #406), merged 2026-07-05, +41/−17 in 6 files. Added hover tooltips to the toolbar icons so each tool explains itself instead of having to be clicked to be understood. - [DataLoom #410](https://github.com/c2siorg/dataloom/pull/410) — "feat: apply-preview workflow for all transformations" (Feature; follows up #405), merged 2026-07-05, +444/−499 in 20 files. A large cross-cutting refactor extending the preview-before-apply workflow to all 12 transformation forms, so no transformation can change the dataset without explicit confirmation. - [DataLoom #383](https://github.com/c2siorg/dataloom/pull/383) — "fix: add preview-before-persist flow for row-reducing transforms" (Fix; addresses #374), merged 2026-06-27, +484/−330 in 19 files. Introduced a preview-before-persist flow for row-reducing transformations, so users see the result before their data is changed rather than after. - [TensorMap #367](https://github.com/c2siorg/tensormap/pull/367) — "fix: resolve JSON serialization error for NaN values in dataset preview" (Fix; closes #366), merged 2026-06-21, +3/−1 in 1 file. Fixed NaN values breaking JSON serialization, which had made the dataset preview fail on any CSV containing empty cells. - [DataLoom #348](https://github.com/c2siorg/dataloom/pull/348) — "fix: resolve strict case-sensitivity in string filtering" (Fix; closes #327), merged 2026-06-14, +26/−8 in 2 files. Made string filtering case-insensitive, so a filter for "cash" no longer silently missed rows containing "Cash" or "CASH". ### Open issues he opened - [DataLoom #601](https://github.com/c2siorg/dataloom/issues/601) — "[RFC] Deterministic operation suggestions, low-memory targets, and optional local AI assistant", opened 2026-10-27. - [DataLoom #537](https://github.com/c2siorg/dataloom/issues/537) — "[Feature] Run slow operations as background jobs with progress reporting", opened 2026-09-30. - [DataLoom #533](https://github.com/c2siorg/dataloom/issues/533) — "[Feature] Lift the 10 MB upload cap safely with streamed upload and sample-based profiling", opened 2026-09-29. - [DataLoom #524](https://github.com/c2siorg/dataloom/issues/524) — "[Feature] Use Parquet for the internal working copy so dtypes persist between requests", opened 2026-09-27. ### Code review - [DataLoom #457](https://github.com/c2siorg/dataloom/pull/457) — "feat: add paginated preview support for transformations" by @SHIVAM-KUMAR-59. Raised two findings on another contributor's paginated-preview PR: an in-flight race in `fetchPreviewPage`, which the author fixed in the PR, and a full re-transform on every page change, which the author agreed with and recorded as a follow-up. - [DataLoom #412](https://github.com/c2siorg/dataloom/pull/412) — "fix(security): bump pyjwt and python-multipart to patched releases" by @anxkhn. Pointed out that the new regression test imports `packaging` without declaring it as a dev dependency. ## Writing Yadunandan Singh has written 11 articles, all on his blog at https://yadunandan-singh.pages.dev/blog/ (first published on Medium, rewritten for the site). Each entry links to the article, gives its publication date and a summary, and links a plain-markdown copy of the full text. Every article in full, in one file: https://yadunandan-singh.pages.dev/llms-full.txt - [Three Signals, One Verdict, and a Payment That Waits](https://yadunandan-singh.pages.dev/blog/veracity-scam-call-detector-architecture/) (published 2026-11-08; topics: System Design, Cybersecurity, Deepfake Detection; full text as markdown: https://yadunandan-singh.pages.dev/blog/veracity-scam-call-detector-architecture.md): In Part 2 of his first-hackathon series, Yadunandan Singh explains how Veracity, the scam-call detector that won second place at CT BuildFest 2.0, works. An Android phone on a real call is connected to a laptop by USB; scrcpy and BlackHole bring the caller's voice into a FastAPI backend that checks three signals at once: the caller's number (libphonenumber and a transparent rule score), the voice (an AI-voice model, now Spectra-0, scoring a 4-second window every second) and the words (Whisper, scam rules in English, Hinglish and Hindi, and a local Llama 3.1 8B that may only cite quotes found in the transcript). A weighted fusion lets any one strong signal raise the risk on its own, and every call ends in a JSON evidence report. He has since taken Veracity into the PayPal AI Hackathon, where a payment gate holds PayPal payments authorized during a risky call until the caller is verified. Lessons from the build are saved for Part 3. - [30 Hours, One Laptop, and a Scam-Call Detector](https://yadunandan-singh.pages.dev/blog/first-hackathon-scam-call-detector/) (published 2026-11-01; topics: Hackathon, Cybersecurity, Deepfake Detection; full text as markdown: https://yadunandan-singh.pages.dev/blog/first-hackathon-scam-call-detector.md): Yadunandan Singh recounts his first hackathon, CT BuildFest 2.0 in Jalandhar, where Veracity, which he coded over 30 hours for his six-member team Coffee to Code, won second place. During a live call, Veracity checks the caller's number, how likely the voice is AI-generated (wav2vec2) and the words (Whisper plus Llama 3.1 run locally) for scam patterns in English, Hindi and Hinglish, warns the user mid-call and builds an evidence report for the 1930 helpline. Part 1 of 2: the build, the mentors' evaluation and lessons on teamwork. - [I Used AI to Predict My University Exam Paper, So I Turned It Into an Open-Source Claude Skill](https://yadunandan-singh.pages.dev/blog/exam-paper-predictor/) (published 2026-10-26; topics: Claude, Open Source, Learning; full text as markdown: https://yadunandan-singh.pages.dev/blog/exam-paper-predictor.md): Yadunandan Singh describes how a last-minute hostel experiment with previous year question papers (PYQs) and Claude produced a mock exam that overlapped the real paper by roughly 90%. He packaged the workflow into exam-paper-predictor, a free, open-source Claude skill for university, board, entrance and professional exams in any country or language. The skill collects 5 to 20 years of past papers, maps section and topic weightage, backtests its own predictions on held-out past years before generating a mock paper, exports ready-to-use PDFs (mock paper, solutions and important topics), and scores its prediction against the real paper afterwards. - [How to Find Your First Open-Source Project in One Evening](https://yadunandan-singh.pages.dev/blog/find-your-first-open-source-project/) (published 2026-07-24; topics: Open Source, GitHub, Career; full text as markdown: https://yadunandan-singh.pages.dev/blog/find-your-first-open-source-project.md): Yadunandan Singh explains the method he used to go from never having opened a pull request to eight merged pull requests in the c2siorg organisation (DataLoom and TensorMap) within forty days. Answer three questions on paper (your strongest language, your domain and your realistic weekly hours), turn them into one targeted AI search prompt, then spend 90 seconds vetting each suggested repository. The second half is an honest, step-by-step account of a first pull request, from local setup to review and merge. - [ChatGPT Failed My Sister’s Live Class, So I Built a Neuro-Symbolic Math Solver](https://yadunandan-singh.pages.dev/blog/neuro-symbolic-math-solver/) (published 2026-05-30; topics: Machine Learning, Computer Vision, Neuro-Symbolic AI; full text as markdown: https://yadunandan-singh.pages.dev/blog/neuro-symbolic-math-solver.md): After watching ChatGPT answer a simple numerical with an essay during his sister's live physics class, Yadunandan Singh built a neuro-symbolic math solver. A React canvas sends a handwritten equation to a FastAPI backend, which normalises the image's polarity, segments each symbol with OpenCV, classifies it with a CNN trained from scratch on MNIST plus 24,000 synthetic operator images, and evaluates the expression deterministically, so the answer is exact rather than generated. - [Stop Vibe Coding: How Elite Developers Actually Use AI](https://yadunandan-singh.pages.dev/blog/stop-vibe-coding/) (published 2026-04-07; topics: AI-Assisted Coding, Software Engineering, Research; full text as markdown: https://yadunandan-singh.pages.dev/blog/stop-vibe-coding.md): Yadunandan Singh reviews what research says about AI-assisted programming: a December 2025 study found that experienced developers keep tight control of AI agents rather than "vibe coding", and METR's randomised trial found developers who felt 20% faster were measured 19% slower. He sets out a seven-step workflow drawn from that research and Addy Osmani's process: spec first, small steps, full context, immediate tests, line-by-line review, a second model's opinion, and regular practice without AI. - [How I Built a Face Authentication App in Django: A Complete ML Learning Journey](https://yadunandan-singh.pages.dev/blog/django-face-authentication-app/) (published 2025-12-22; topics: Machine Learning, Django, Learning; full text as markdown: https://yadunandan-singh.pages.dev/blog/django-face-authentication-app.md): Yadunandan Singh describes building a Django web app that users log in to with their face, using the eigenfaces method: greyscale 100 × 100 face images are flattened, centred on the mean face and decomposed with SVD in NumPy, and logins are matched by Euclidean distance with a rejection threshold. Because one photo per user wasn't enough to train on, OpenCV generates eight synthetic variations of each sign-up photo. The post also sets out his method for learning a new framework through a project without copying AI-generated code. - [Should You Be Scared of Agentic AI? Here’s the Reality](https://yadunandan-singh.pages.dev/blog/should-you-be-scared-of-agentic-ai/) (published 2025-12-05; topics: Agentic AI, AI-Assisted Coding, Career; full text as markdown: https://yadunandan-singh.pages.dev/blog/should-you-be-scared-of-agentic-ai.md): Yadunandan Singh explains what agentic AI is (systems that plan, act across tools and check their own results) and reports on testing Google Antigravity, launched with Gemini 3 in November 2025, by generating a portfolio site with React, Django and OTP authentication. His conclusion: agents are excellent for prototypes and internal tools but weak for long-lived production code, beginners should not use them to write their main code, and AI rewards developers with strong fundamentals rather than replacing them. - [Don’t Be a Dumb Computer: SVD Explained With a YouTube Recommender](https://yadunandan-singh.pages.dev/blog/svd-explained-youtube-recommender/) (published 2025-11-17; topics: Linear Algebra, Machine Learning, Recommender Systems; full text as markdown: https://yadunandan-singh.pages.dev/blog/svd-explained-youtube-recommender.md): Yadunandan Singh explains Singular Value Decomposition by working a small example by hand: a 3 × 4 matrix of liked and disliked videos decomposes into concept strengths of 1.41, 0.92 and 0.20, and truncating to k = 2 keeps "STEM" and "entertainment" while discarding noise. He then shows how his YouTube recommender uses TF-IDF and scikit-learn's TruncatedSVD to place videos on that concept map, builds a taste profile as the average liked vector minus the average disliked one, and ranks candidates by cosine similarity. - [I Stopped Solving Problems and Built a Tool Instead: How a Visualizer Taught Me Real Linear Algebra](https://yadunandan-singh.pages.dev/blog/linear-algebra-transformation-visualizer/) (published 2025-10-14; topics: Linear Algebra, Mathematics, Python; full text as markdown: https://yadunandan-singh.pages.dev/blog/linear-algebra-transformation-visualizer.md): Yadunandan Singh built an interactive visualiser that shows what a 2 × 2 matrix does to the unit square, and uses it to explain linear algebra through hand-worked examples. A shape is stored as a 2 × 5 matrix of vertices, so one multiplication transforms every point; a matrix's columns show where the basis vectors land; the determinant is the factor by which area scales, negative for reflections and zero when space collapses onto a line; and eigenvectors are the directions that only stretch. - [How to Build Your Own Image Recognition Model as Your First AI Project](https://yadunandan-singh.pages.dev/blog/first-image-recognition-model/) (published 2025-09-24; topics: Machine Learning, Computer Vision, Beginners; full text as markdown: https://yadunandan-singh.pages.dev/blog/first-image-recognition-model.md): Yadunandan Singh documents his first hands-on AI project: a convolutional neural network trained on CIFAR-10 with TensorFlow and Keras and served through a Flask web app. The guide covers how images become 32 × 32 × 3 tensors, normalisation, data augmentation, three Conv2D and pooling blocks, dropout, and the compile and fit settings, plus a 1,563-epoch, eight-hour experiment in which strong augmentation and dropout kept validation accuracy improving instead of overfitting. ## Pages - [Full profile (plain HTML)](https://yadunandan-singh.pages.dev/profile/): everything below in one static page. - [about.md](https://yadunandan-singh.pages.dev/about.md): this profile as plain markdown. - [Portfolio home](https://yadunandan-singh.pages.dev/): projects, writing and open-source history. - [Experience](https://yadunandan-singh.pages.dev/about/): open-source record and skills. - [Projects](https://yadunandan-singh.pages.dev/library/): project list with repository links. - [Blog](https://yadunandan-singh.pages.dev/blog/): long-form articles, each with its own page. - [llms-full.txt](https://yadunandan-singh.pages.dev/llms-full.txt): this file plus the full text of every article. - [Blog RSS feed](https://yadunandan-singh.pages.dev/blog/rss.xml): every article with its summary and full text. - [Photographs](https://yadunandan-singh.pages.dev/photos/): official photographs.