Aryan Kumar

web developer in the UK, learning AI

Query me.I'm learning how this works.

I'm Aryan Kumar, a web developer in the UK. I build whole products in TypeScript, from the database to the iOS build, and now I'm teaching myself to build AI: RAG and agents.

See the workMessage me on LinkedIn

aryan-kumar23 chunksindex built 16 Oct 2026
web workAI learningmeShyft: a UK shift and take-home pay app, on the web and in iOS review.ShyftShyft's tax engine matches all 1,000 entries of HMRC's Table A.HMRC 1,000/1,000Shyft counts shifts in UK clock time, so a 22:00 to 06:00 shift is 7 hours on the night the clocks go forward and 9 on the night they go back.clock-change mathsShyft: 592 commits since August 2025, built solo.592 commitsCODEMYBM UK: my first web-dev client, a BMW and MINI coding workshop in Brierley Hill.first clientCODEMYBM UK: rebuilt in 2 days, Lighthouse performance 90 on mobile and 100 on desktop.2 days, Lighthouse 90CODEMYBM UK: a 3D wireframe BMW M4 CSL, cut from a 70 MB model to 3 MB.70 MB to 3 MBEstateLink: property management for UK landlords, built over 7 months.EstateLinkEstateLink: live dashboard numbers pushed over logged-in websockets.live websocketsStack I build with: TypeScript, React, Vite, Next.js, Supabase, Postgres, MongoDB.my stackI build products alone, from the database schema to App Store submission.built aloneMythos: my first LLM call. Type a chore, get back a fantasy quest with XP.first LLM callMythos checks the model's JSON and clamps its XP between 10 and 150.XP clamped 10 to 150Learning now: RAG, getting a model to answer from my own documents.RAG (learning)Learning now: AI agents.agents (learning)This map is a toy: keyword scoring, not real embeddings. Yet.this map is a toyFinal-year project at the University of Birmingham, 2024 to 2025.Birmingham, 2024-25Peak commit hour: 23:00. 113 of my 990 commits.peak hour 23:00261 of my commits landed between midnight and 5am.261 after midnightMy busiest day is Monday: 231 commits.MondaysI write British English and I don't invent numbers. Every figure here has a source.no invented numbersBased in the UK.UKGitHub: AryanKumar02.GitHub

Keyword scoring for now, not real embeddings. My real RAG is in progress.

    All 23 chunks

    web work

    AI learning

    me

    Retrieved: things I built

    Four results, best first. Each one opens the full story, with sources.

    1. My product, live on the web, iOS in review2025 to now

      Shyft

      Log your shifts and see your real take-home pay after UK tax, National Insurance, student loan and pension.

      Its tax engine matches all 1,000 entries of HMRC Table A.Shift-Tracker client/src/lib/__tests__/hmrcTableA.test.ts

    2. Client work, my first web-dev client2026

      CODEMYBM UK

      A full rebuild for a BMW and MINI coding workshop in Brierley Hill, with a "check my car" tool and a 3D wireframe M4 CSL.

      Rebuilt in 2 days. Lighthouse performance 90 on mobile.CodeMyBM site/QA-REPORT.md, section 2; git log

    3. My product2025

      EstateLink

      Property management for UK landlords: properties, tenants, leases and a live dashboard.

      148 commits over 7 months, built alone.EstateLink git log, May to December 2025

    4. Final-year project, University of Birmingham2024 to 2025

      Mythos

      Type a chore and an LLM turns it into a fantasy quest with XP, levels and streaks.

      My first LLM call, with its output checked before it reaches the screen.Project-RPG backend/controllers/questController.js

    Learning now

    RAG and AI agents. I'm at the start: no shipped RAG system yet, and this site's map is a keyword toy rather than real embeddings. When I build the real thing, it'll show up here first.

    More about me