about_me.txt — Notes
YOIHEN ELANGBAM Data Analyst · Bengaluru, India status: open to data & analytics roles > I turn messy operations data into decisions people act on. EDUCATION Manipal Institute of Technology BTech, Data Science Engineering · 2026 CGPA 7.24 · DSA, ML, Big Data, DBMS LEADERSHIP Founder, North Eastern Students Association, Manipal (40+ members) Sergeant at Arms, Amazon Bangalore Toastmasters English Teacher, ENGin RECOGNITION Dragon 100 Delegate · Hong Kong 2026 Phoenix Fellowship · Hong Kong 2025 Delegate for India, HPAIR · Bangkok 2024
Ln 1, Col 1
Experience
Address
C:\Yoihen\Experience
Name
Role
Dates
    4 objects · click one to view details
    Projects
    Address
    C:\Yoihen\Projects
    2 objects
    CookNoBook — Web Browser
    Address
    https://cooknobook.com
    Go ↗
    CookNoBook recipe grid screenshot

    CookNoBook

    Recipes as a visual grid: ingredients run down the left and each step spans the ingredients that go into it, so you see the whole dish at a glance. Claude writes any dish on demand, and it's built to stay cheap enough to run on a free tier.

    CHAPTER 1 Paragraphs are hard to cook from You lose your place mid-step and miss what needs prepping in parallel.
    CHAPTER 2 So I made recipes visual Each recipe is a tree (ingredients feed steps, steps feed the dish) shown as a grid, flow graph or cards.
    CHAPTER 3 AI made any dish possible Claude writes recipes straight into that structure, branches and all. TheMealDB is the fallback.
    CHAPTER 4 Then: affordable and abuse-proof A public AI app costs money per request, so I engineered the costs and the abuse out.
    Keeping AI costs under control
    • Library recipes first, then saved ones; Claude is only called when nothing else has the dish.
    • Fuzzy dish matching ("lasagna" = "lasagne") so the same dish is never paid for twice.
    • Library pre-filled with Anthropic's half-price Batch API.
    • Per-visitor rate limit, daily caps and an hourly circuit breaker protect the budget.
    • Every call logs its token use; /api/health reports cost per recipe.
    Bot & abuse protection
    • Cloudflare Turnstile guards anything that costs money; passing it issues a signed HttpOnly cookie.
    • Limits apply per IP and per pass, so switching addresses doesn't get around them.
    • Slow and flooding connections are capped.
    Under the hood
    • Self-fixing library: after 3 visitor reports, an AI recipe is pulled and rewritten with the complaints included, at most once a week per dish.
    • SEO without a framework: the server writes titles, share previews and schema.org Recipe data into each page, plus a sitemap; no JavaScript needed to index.
    • Few dependencies: Python backend whose only package is anthropic; plain JavaScript, no build step; self-hosted fonts, no calls to Google.
    • Storage that survives restarts: Upstash Redis keeps saved recipes through deploys on Render's free plan; rare recipes expire, popular ones are promoted to the library.
    Headline features
    Three views of one recipeGrid, flow graph and step-by-step cards.
    Any dish on demandClaude writes it in structured form, with branching steps.
    Cooking modeTimers, unit conversion, read-aloud steps, screen kept awake (Wake Lock).
    AccessibleDark/light themes, Atkinson Hyperlegible easy-reading mode, adjustable text size.
    Python · anthropic · vanilla JS · Upstash Redis · Cloudflare Turnstile · Render
    Done
    Art Portfolio — Web Browser
    Address
    https://artportfolio-rh8r.onrender.com
    Go ↗
    Art portfolio cover: a glowing PORTFOLIO. title surrounded by the section names

    Art Portfolio

    A single-page portfolio for my album packaging, graphic design, photography, film and writing, designed and built from scratch in plain HTML, CSS and JavaScript, with three.js for the 3D pieces and no build step. It is hosted as a static site that redeploys on every push to GitHub.

    3D packaging
    • Every vinyl, CD and cassette is drawn live with three.js, lit by a studio of softboxes and casting its own shadow.
    • Records slide out and spin with anisotropic highlights, cassette reels turn, and stacked sleeves come apart on hover and reassemble when the cursor leaves.
    • One shared renderer draws all nine pieces, and nothing is drawn while they are still.
    From mock-ups to textures
    • With no print files, a Python and OpenCV pipeline lifts each face from the photos and video loops.
    • Spinning discs are rebuilt by unspinning the frames and taking the median, which removes the studio highlights.
    • For the angled stacks, the camera is recovered from a sleeve's square, so the 3D model lines up with the photo to within half a pixel.
    Interface and motion
    • Design prints: the covers become glossy cards that flip over to the next piece, changing shape while edge-on.
    • Cover: a SUPERHOT-style title chant in WebGL, with a fragment-shader slam that knocks the surrounding text outward.
    • Seamless scrolling: full-screen glass cards whose backdrops and tints cross-fade on fixed layers, changing only opacity.
    • Persona 5-style navigation: a cut-in that shoots the chosen chapter word, aimed by solving the barrel's angle.
    • Film reviews from Letterboxd, with the full review shown on hover.
    Performance
    • Heavier parts lazy-loaded, and three.js cut down to 135 KB compressed.
    • A trimming pass took peak memory from about 2.25 GB to 1.4 GB.
    • The 3D pieces hold 60 fps.
    JavaScript · three.js · WebGL · Python · OpenCV · Web design
    Done
    Toolkit Properties
    Installed components:
    impact.exe — Performance
    $42M
    Cost defect foundEntitlement-capture defect in Amazon US Retail.
    -90%
    Report time3–4 days to 1–2 hours via an Ops Variable chatbot.
    95%
    Model accuracyRandom forest predicting funnel completion at Fidelity.
    -95%
    Prompt tokensCookNoBook's JSON-LD extraction step.
    Processes: 4 · Uptime: since 2024
    New message
    ?
    Send Yoihen an email?
    Yes
    resume.pdf — Document Viewer
    Yoihen_Elangbam_Resume.pdf · 1 page
    Résumé of Yoihen Elangbam
    Loading document…
    Page 1 of 1Open in new tab
    Welcome
    i
    Hi, I'm Yoihen, a data analyst. Click any icon on the desktop to explore my work, or use the start menu below.

    Wallpaper: © Vyacheslav Argenberg, CC BY 4.0