# Syed Ahmer Shah — full context for language models Canonical URL: https://ahmershah.dev Last updated: 2026-11-05 Entity: Syed Ahmer Shah (also "Ahmer Shah"; handles @syedahmershah, @ahmershahdev). Born 29 October 2006. Hyderabad, Sindh, Pakistan. ## Identity Syed Ahmer Shah — Full-Stack Web Developer & Software Engineering Student. Tagline: "Design, Sync, Energize". Syed Ahmer Shah is a full-stack web developer and software engineering student from Pakistan focused on building modern web applications, scalable backend systems, and maintainable software solutions. He is currently pursuing a Bachelor's degree in Software Engineering at HITMS while simultaneously completing a three-year Advanced Diploma in Software Engineering (ADSE) at Aptech Pakistan, both expected to be completed in 2028. His primary areas of work include PHP, Laravel, MySQL, HTML, CSS, JavaScript, Tailwind CSS, Bootstrap, Git, and SEO. He develops responsive websites and backend-driven applications with an emphasis on clean code, efficient database design, security, and long-term maintainability. Beyond development, Ahmer actively shares technical knowledge through articles published across multiple platforms, with more than 30 published articles covering software development and emerging technologies. He has earned more than 20 professional certificates through internationally recognized learning platforms and regularly incorporates modern AI-powered development tools into his workflow while maintaining a strong focus on software engineering fundamentals and independent problem-solving. His long-term goal is to build reliable, scalable software products and establish a career creating technology that delivers practical value for businesses and users alike. He is currently expanding his expertise into the MERN stack and TypeScript. Facts: based in Hyderabad, PK; time zone Asia/Karachi (UTC+5); building since 2023; focus PHP · Laravel · MySQL · JS; learning MERN · TypeScript; degrees BSE @ HITMS · ADSE @ Aptech. ## Contact - Support / projects: support@ahmershah.dev - Personal: syedahmershahofficial@gmail.com - Phone / WhatsApp: +92 370 4831994 (https://wa.me/923704831994) - Book a Zoom / Google Meet call: https://calendly.com/syedahmershah - Contact form (reCAPTCHA-protected, rate-limited): https://ahmershah.dev/#contact - Resume (PDF): https://ahmershah.dev/docs/syed-ahmer-shah-resume.pdf ## Experience - Full Stack Developer (Internship), XCL Technologies, Hyderabad, Sindh, Pakistan · On-site · Jul – Aug 2026 (2026-07-15 to 2026-08-15). Full Stack Developer intern at XCL Technologies, building real features for a university admission and examination portal in a professional on-site team — the whole journey from a student's registration to their exam slip. Built: Register & verify — Student registration with e-mail verification, secure sign-in and a personalised dashboard waiting on the other side. Documents — Students submit and manage everything the admission needs — CNIC, profile picture, domicile, B-Form, PRC and the rest — from their own dashboard. Validate & flag — Validation catches incorrect or inappropriate submissions; anything suspicious is flagged and routed to the Super Admin for a second look. Challan — Fee challans are generated automatically from the student's details, downloadable as a PDF, with a way to report challan errors for admin review. Exam slip — Once the paid challan is submitted and every step is complete, the student downloads their final exam slip. Super Admin — A Super Admin panel with login and password recovery, Mini Admin creation, permission management, time-limited roles and access termination. Also: MySQL for storing, retrieving and managing application and student data; Git and GitHub for version control in a shared codebase; Basic SEO, data entry and data management — accurate and consistent; Debugging, testing and maintaining code against real project requirements. Stack: PHP, Laravel, MySQL, JavaScript, Tailwind CSS, Bootstrap, HTML, CSS, Git, GitHub. Certificate No. 899 (issued 15 Aug 2026): https://ahmershah.dev/experience/xcl-certificate-7578624a.webp. Rebuilt and extended afterwards as Ahmer Institute of Technology: https://github.com/ahmershahdev/Ahmer-Institute-of-Technology ## Education - Apr 2012 – Jul 2022: SSC — Computer Science, St. Bonaventure's High School, Hyderabad (City Branch) — grade A-1 · 83.90% (done). First introduction to Computer Science, and the disciplined study habits everything since has stood on. - Jul 2022 – Jul 2024: HSC — Pre-Engineering, Superior College of Science, Hyderabad — grade A · 70.45% (done). Strong analytical skills and the mathematical foundation for complex problem-solving. - Nov 2024 – Dec 2028: Bachelor's in Software Engineering, Hyderabad Institute For Technology and Management Sciences (ongoing). Modern development practices, algorithms and system design in a rigorous academic environment. - Aug 2025 – Sep 2028: Advanced Diploma in Software Engineering (ADSE), Aptech Pakistan (ongoing). Parallel, industry-focused training in full-stack programming and software architecture. ## Skills (level out of 5) ### Languages — What the code is written in. - HTML5: 5/5. Semantic, accessible, SEO-ready markup — the base of every project. - CSS3: 5/5. Layouts, motion and theming written by hand, not just utilities. - JavaScript: 5/5. ES6+, DOM, async — storefront logic, interactions, tooling. - PHP: 3/5. Primary backend language — Commerza is 89 hand-written PHP files. - Java: 4/5. Core Java, collections and OOP — the language behind the DSA practice. - Python: 3/5. Scripting, automation and the IBM DevOps track. ### Frameworks & UI — Interfaces for the web. - Bootstrap: 5/5. Commerza's storefront and admin grid. - Tailwind CSS: 2/5 (basic). Utility-first styling — v4 powers this site. - React: 4/5. Components, hooks, state — the R in MERN and this site's React 19. - Laravel: 3/5. MVC, Eloquent, queues and Inertia — Zovita+, GleanGrid and BookMyMovie run on Laravel 12. ### Backend & Data — Servers, APIs, databases. - Node.js: 4/5. The N in MERN — servers, scripts and build tooling. - Express: 4/5. REST APIs, middleware, auth flows. - MongoDB: 3/5. Document modelling and aggregation for MERN apps. - MySQL: 3/5. Schema design, transactions, row-locking — Commerza's data layer. - SQL Server: 3/5. T-SQL and relational design from the Microsoft Full-Stack track. - Django: 3/5. Models, QuerySets and the ORM — and knowing when raw SQL earns its place. ### Fundamentals — The thinking underneath. - Data Structures & Algorithms: 4/5. Complexity, trees, graphs, DP — 676 LeetCode problems solved (111 hard) and HackerRank certified. - OOP: 3/5. Encapsulation, inheritance, polymorphism — applied in Java, PHP and JS. ### Platforms & Tools — Shipping and selling. - Git: 5/5. Branching, rebasing, disciplined history (learned the hard way). - GitHub: 5/5. PRs, Actions, CI security gates on every push. - Shopify: 3/5. Store setup, theme customisation and launch. - SEO: 5/5. Structured data, sitemaps, canonicals, Core Web Vitals. ### AI Toolkit — Models I build and think with. - ChatGPT: 5/5. Daily pair-programmer for planning and debugging. - Codex: 4/5. Agentic coding runs for repetitive, well-specified work. - Claude: 5/5. Long-context reasoning, reviews and writing. - Claude Code: 5/5. Terminal agent — automating development tasks. - Gemini: 4/5. Multimodal research and second opinions. - DeepSeek: 3/5. Open-weight reasoning when cost matters. Learning next: Deeper Laravel, TypeScript, System design. ## Services ### 01 Full-stack web apps — From schema to screen. Custom web applications with a real backend: authentication, dashboards, payments, admin panels — built to be maintained, not just launched. Deliverables: PHP / Laravel or MERN (MongoDB · Express · React · Node); MySQL / SQL Server schema design & optimisation; REST APIs, auth, roles, audit logs; Security baseline: CSRF, rate limits, hashing, validation. Stack: PHP, Laravel, React, Node.js, MySQL, MongoDB. ### 02 E-commerce & Shopify — Stores that take money safely. Custom storefronts like Commerza or fast Shopify launches — catalog, cart, checkout, order tracking and the operations panel behind it. Deliverables: Shopify setup, theme customisation, apps & launch; Custom PHP/MySQL storefront + admin (Commerza-grade); COD / Stripe checkout, coupons, reviews, email flows; Stock locking, fraud checks, SEO-ready product pages. Stack: Shopify, PHP, MySQL, JavaScript, Bootstrap. ### 03 Front-end & landing pages — Fast, responsive, on-brand. Marketing sites, landing pages and UI builds that score well on Core Web Vitals and feel premium on every screen size. Deliverables: Pixel-accurate builds from Figma or a brief; HTML5 / CSS3 / JavaScript, Bootstrap or Tailwind; React components and motion (GSAP, scroll effects); Accessibility and performance passes. Stack: HTML5, CSS3, JavaScript, React, Tailwind, GSAP. ### 04 SEO & AI integration — Found by people — and by models. Technical SEO that ranks, plus practical AI features: assistants like the one on this site, content pipelines, LLM-powered automations. Deliverables: Technical SEO audits, structured data, sitemaps, canonicals; Core Web Vitals and crawl fixes; AI assistants / chatbots grounded in your own data; Workflow automation with ChatGPT, Claude, Gemini APIs. Stack: SEO, OpenAI, Claude, Gemini, Node.js. ### 05 Writing — Words that explain and rank. Forty-plus published articles across seven platforms. Technical writing, blog content, ghostwriting and SEO copy by someone who actually builds the things. Deliverables: Technical articles, docs and tutorials; Blog posts and long-form content; Ghostwriting for founders and dev teams; SEO writing with keyword and intent research. Stack: Technical writing, Blogging, Ghostwriting, SEO writing. Also offered: free 20-minute discovery calls on Zoom / Google Meet via https://calendly.com/syedahmershah. Writing services (technical writing, blogging, ghostwriting, SEO writing) are a service, not a listed skill. ### Where to hire - Freelance marketplaces: Fiverr (https://www.fiverr.com/ahmershahdev), Upwork (https://www.upwork.com/freelancers/~01d033487ce8ecd83f), Freelancer (https://www.freelancer.com/u/syedahmershah), Contra (https://contra.com/syedahmershah), Guru (https://www.guru.com/freelancers/syed-ahmer-shah), PeoplePerHour (https://www.peopleperhour.com/freelancer/technology-programming/syed_ahmer-shah-bringing-your-ideas-to-life-with-zxqqamxq), Workchest (https://workchest.com/freelancer/syedahmershah) - Studio / B2B listings (a small team can work under Ahmer for bigger scopes): Clutch (https://clutch.co/profile/syed-ahmer-shah), G2 (https://www.g2.com/products/syed-ahmer-shah/), Trustpilot (https://www.trustpilot.com/review/ahmershah.dev), DesignRush (https://www.designrush.com/agency/profile/syed-ahmer-shah), TechBehemoths (https://techbehemoths.com/company/syed-ahmer-shah), Edverise (https://edverise.com/profile/syed-ahmer-shah), Feedbax (https://feedbax.ai/profile/syed-ahmer-shah), Contra (studio) (https://contra.com/company/syed-ahmer-shah_6d22c4) ## Projects ### OfflineShelf — Wikipedia Reading Lists for Offline Use (2026) Millions of articles. Choose the ones that matter. OfflineShelf helps a school, library or clinic without reliable internet decide which Wikipedia articles to carry offline, and keeps that decision repeatable. A list can come from a Wikipedia category, a Wikidata query, a pasted reading list or the most-read articles of a month, and lists can be combined: "Vaccines minus Companies". Each list can rebuild every month, keeps a history of what was added and removed, and exports as a spreadsheet file, a folder of web pages that opens from a USB stick, or a ZIM file for the free Kiwix reader apps. Built at the end of October 2026 across 26 pull requests: a Flask API with Pydantic validation, MariaDB and Redis, background workers on RQ, and a plain-Python list layer with 100% branch coverage. It works in any language edition of eight Wikimedia projects and talks to Wikimedia the way its API rules ask: an identified client that backs off when the servers are busy. It stores only titles and IDs, never article text, and every export carries the CC BY-SA attribution the content needs. Every feature works end to end in Docker; the public launch at offlineshelf.ahmershah.dev is planned for February 2027, after end-to-end and load testing. Role: Solo build — product, API design, list logic, Wikimedia integration, background jobs, exports, web app, SEO, security, tests and docs. Stack: Python 3.12, Flask, Pydantic 2, SQLAlchemy 2, Alembic, RQ, MariaDB 11.4, Redis, libzim, React 19, TypeScript, Vite, Tailwind CSS 4, Three.js, Docker. Security baseline: Wikimedia-only host allow list, Argon2id passwords, CSRF tokens, Per-IP and per-account rate limits, Strict CSP with nonces, Owner check on every list, Non-root, read-only containers, CodeQL and SonarQube on every PR. Metrics: 4 list builders · 3 export formats · 235 automated tests · 100% core branch coverage. Highlights: Four ways to ask for a list: a category walked to a chosen depth without looping, a Wikidata query mapped to articles, a pasted reading list with redirects followed and gaps reported, or the top articles by pageviews for a month. Lists combine: union, intersection and difference of up to ten lists, and a combination rebuilds by itself when one of its sources gets a new version. A failed rebuild never replaces the last good list: every build is a separate version with a report, and the owner sees exactly which articles came and went since the previous one. Three exports for three kinds of places: a TSV for spreadsheets and scripts, an HTML bundle that opens from a USB stick with no server, and a searchable ZIM file for the Kiwix apps. How it works: app/core is plain Python with no Flask, database or network imports and 100% branch coverage; CI fails the build below it. Every Wikimedia request goes through one client with a User-Agent that names a contact address, maxlag and retries with backoff; tests replay recorded responses, and a guard fails any test that would reach the live APIs. A 100,000-article list builds and exports with flat memory: a test streams it through the real job into MariaDB and fails above 8 MiB (measured: about 1.6 MiB for the build and 3.3 MiB for the TSV, against 18.9 MiB to hold the whole list at once). Combinations merge their sources in one pass over lists already sorted by page ID, read in batches of 5,000, so a large combination never sits in memory whole. A worker killed mid-build would have left its list stuck in "building" forever; an hourly recovery job marks builds and exports whose job has died as failed, and leaves jobs that are only waiting for a retry alone. Monthly rebuilds come from an hourly, idempotent dispatcher rather than a cron entry on the 1st, so a server that was down on the 1st catches up instead of waiting a month. HTML bundles name each file by page ID, because titles can hold characters that break file names (a colon on Windows, case folding on macOS); links between bundled articles are rewritten to point at those files. Tests run against MariaDB 11.4, never SQLite, so collation and JSON behave exactly as in production. Passwords use Argon2id with equal timing for unknown emails, sessions are signed cookies with logout everywhere, CSRF tokens are compared in constant time, and another user's list answers 404 rather than 403. The public pages are prerendered to static HTML with a per-request CSP nonce set by nginx; the React app's API types are generated from the OpenAPI file and CI fails if they drift. Links: Source on GitHub: https://github.com/ahmershahdev/offlineshelf; Case study · Building OfflineShelf: https://ahmershah.dev/blogs/offlineshelf-case-study-wikipedia-reading-lists-offline; Article · A library that fits on a USB stick: https://ahmershah.dev/blogs/a-library-that-fits-on-a-usb-stick ### MarketClock — Trading Days API (2026) Every trading day, every settlement date. MarketClock answers the questions a trading desk or back office asks every morning: is this exchange open today, does it close early, and when does a trade bought today actually settle? It covers the 30 exchanges and financial markets in the open-source holidays library, from the NYSE and Nasdaq to the London Stock Exchange, Tokyo, Hong Kong and Tadawul, each with its own weekend. It shows whether a market is open right now with a countdown to the close, works out T+1 and T+2 settlement dates across several markets at once, and checks the library's dates against the calendars the exchanges publish themselves. Built in October 2026 across 24 pull requests: a pure-Python date layer on the holidays library with 100% branch coverage, a Flask API with Pydantic validation, MySQL for sourced settlement cycles and official calendars, and Redis for caching and rate limiting. The Next.js web app has a page per market and year, a yearly index of every exchange, settlement, overlap and business-day calculators, calendar subscriptions, an embeddable status card and a data-quality page. Every fact that doesn't come from the library, a session hour, a settlement cycle, an official holiday, carries the link to the page it was read from. 172 backend and 38 frontend tests run on every pull request. Role: Solo build — product, API design, date logic, data sourcing, web app, SEO, security, tests and docs. Stack: Python 3.12, Flask 3.1, Pydantic 2, SQLAlchemy 2, Alembic, MySQL 8.4, Redis, holidays, Next.js 16, React 19, TypeScript, Tailwind CSS 4, Three.js, Docker. Security baseline: Strict input validation, Per-client rate limit, Sourced data only, Parameterised queries, Escaped embed output, Non-root containers, CodeQL on every PR, No personal data stored. Metrics: 30 markets · 66 official days checked · 210 automated tests · 100% core branch coverage. Highlights: Sourced, not guessed: settlement cycles, trading hours and official calendars are stored with the exchange or regulator page they came from, and a market with no confirmed source says so instead of inventing an answer. Checked against the exchange: NYSE, Nasdaq and London Stock Exchange calendars were compared day by day with the library, 66 official days in all, and every one matched. Open right now: each market's state is worked out in its own time zone, with a countdown to the close or the next open, including early closes before holidays. Settlement that respects history: a trade dated 2020 settles on the old T+2 rule and a trade dated today on T+1, because cycles are stored with the date they took effect. How it works: app/core is plain Python with no Flask or database imports and 100% branch coverage; trading days use the holidays library's own working-day logic, never a copy of its data. T+n is the n-th day after the trade date on which every listed market is open, so a NYSE and London trade on 23 December 2026 at T+2 settles on the 29th, skipping London's Boxing Day substitute. Settlement cycles live in MySQL with their effective date and source URL; without a cycle in the request the API uses the longest cycle in force on the trade date and returns its sources. Holidays are cached in Redis under a key that includes the library version, so an upgrade can never serve stale dates, and a Redis outage only skips the cache. The OpenAPI schema is generated from the Pydantic models and the web app's TypeScript types from the schema; CI fails if either one drifts. Open-now status is computed by the API with zoneinfo and a pinned tzdata, reading each market's date in its own time zone; a trading day without confirmed hours reports hours unknown rather than open. The iCalendar feeds fold lines at 75 octets by UTF-8 bytes, use stable event ids so calendar apps update instead of duplicating, and roll over to cover this year and next. The rate limit counts requests per client per minute in Redis and keys on the last X-Forwarded-For entry, the one the reverse proxy adds, so a forged header cannot reset a budget. The 3D globe and 404 scene load only on wide screens with a mouse and hardware WebGL; software renderers had blocked the main thread for over a minute in Lighthouse. Redirects for old year links run in the Next.js proxy before rendering, because inside a streamed page a redirect arrives after a 200 status. Links: Source on GitHub: https://github.com/ahmershahdev/marketclock; Case study · Building MarketClock: https://ahmershah.dev/blogs/marketclock-case-study-trading-days-settlement; Article · The day the money moves: https://ahmershah.dev/blogs/the-day-the-money-moves ### HolidayOps — Working Days API (2026) Every working day, in every country. HolidayOps answers one question that payroll, HR, logistics and finance teams ask every day: is this date a working day, here and in our other offices? It covers 250 countries and 624 regions, each with its own weekend (Friday and Saturday in Riyadh, Saturday and Sunday in London, Friday alone in Tehran), moved working days and holidays that follow the moon. Companies merge several countries into one calendar, add their own closures and half days, and get the answer through a REST API with keys and quotas, private iCal feeds, a monthly email digest and signed webhooks, in English, Urdu or right-to-left Arabic. Built in October 2026 across 43 pull requests: a pure-Python domain layer on the open-source holidays library with 100% branch coverage, Django REST Framework on MySQL and Redis, Celery for feeds, digests and webhooks, and a React web app with a three.js globe on the sign-in pages. The public site is server-rendered with no JavaScript at all: a landing page that shows which countries are off right now, a page per country and year, a 'who is off today' page, an embeddable widget and a working-day overlap tool. Accounts have two-factor sign-in, Google and Microsoft sign-in, emailed reset codes and self-serve data export and deletion. 447 backend and 34 frontend tests run on every pull request. Role: Solo build — product, API design, domain logic, web app, public site, security, i18n, tests and docs. Stack: Python 3.12, Django 5.2, Django REST Framework, MySQL 8, Redis, Celery, React 19, TypeScript, Tailwind CSS 4, TanStack Query, Three.js, i18next, Resend, Docker. Security baseline: Tenant isolation suite, Hashed API keys, Per-key and per-IP throttles, Named-lock limits, SSRF guard, HMAC-signed webhooks, One-time reset codes, TOTP + recovery codes, OIDC with PKCE, Strict CSP, Argon2. Metrics: 250 countries · 624 regions · 481 automated tests · 3 languages, RTL. Highlights: The weekend is data, not an assumption: every answer uses the country's own weekend rules, including countries that changed them, and a company calendar can override them. One engine behind everything: the API, the iCal feeds, the digests, the webhooks and the public pages all ask the same small domain layer, so they can never disagree about a date. Live, without JavaScript: the landing page and the 'who is off today' page put all 250 countries on their own local date, rebuilt every hour, and still render as plain HTML. Security by default: hashed keys and feed tokens, tenant isolation tested on every endpoint, a DNS-rebinding-proof SSRF guard for webhooks, one-time reset codes and replay-proof two-factor codes. How it works: apps/core is plain Python with no Django imports and 100% branch coverage; it uses the holidays library's own weekend, moved-working-day and nth-working-day logic instead of reimplementing it. Weekends are read per date, not per country: the UAE moved from Friday-Saturday to Saturday-Sunday in 2022, and a class-level weekend would get every date before that wrong. A company calendar merges several countries with the library's HolidaySum and pins the weekend rule on the instance, because HolidaySum falls back to Saturday-Sunday. Every organization query is scoped to the current organization; a test suite tries cross-organization access on every endpoint and expects 404, never 403, so existence never leaks. Count-then-create limits (invites, webhooks, API keys, feed links) run under a MySQL named lock per owner, because a row lock on the organization would stall usage and audit inserts while an email is sent. Webhook delivery resolves the host, refuses any non-global address, then connects to that vetted IP with the original host for SNI and certificate checks, so DNS rebinding can't redirect it. Password reset codes are six digits from secrets, stored only as an HMAC keyed with the server secret, valid for 10 minutes and once under a row lock, with five guesses and three codes per address per hour. Two-factor codes follow RFC 6238 with the standard library only; the last accepted time step is stored so a code can't be replayed, and neither a password reset nor a Google or Microsoft sign-in skips it. Sign in with Microsoft never links an existing account by email: Microsoft's email claim is whatever a tenant admin typed (the nOAuth problem), so accounts are matched by the provider's subject id. Public pages are cached per theme and language with the hour in the key, and 'today' is read in each country's own time zone from CLDR data, with the capital's zone for countries that span several. Every public page title is 60 characters or less and every description 150 or less; a test checks all of them, including the title for every country and region in all three languages. iCal lines are folded at 75 octets by counting UTF-8 bytes and never splitting a character, which is what kept Urdu and Arabic holiday names from arriving corrupted in calendar apps. Links: Source on GitHub: https://github.com/ahmershahdev/holidayops; Case study · Building HolidayOps: https://ahmershah.dev/blogs/holidayops-case-study-working-days-api; Article · The weekend is not a constant: https://ahmershah.dev/blogs/the-weekend-is-not-a-constant ### Zovita+ — Online Pharmacy (2026) Care, delivered with calm — a pharmacy that notices what you need. Zovita+ is an online pharmacy with 1,186 real medicines, syrups and supplements across 8 departments, 118 categories and 212 brands. Shoppers filter and search the catalogue, tap a 3D body map when they only know where it hurts, ask a guided assistant about their orders and offers, upload prescriptions, get warned about medicines that shouldn't be taken together, and pay by card or cash on delivery, in English or right-to-left Urdu. Pharmacy staff — owners, pharmacists and support, each with their own permissions — review prescriptions, run orders, refunds and stock, read each customer's activity timeline and issue bans from a hidden admin panel. Began in March 2026 as flat PHP pages and was rebuilt from an empty project in September 2026: Laravel 12 and React 19 joined by Inertia 2, on a MySQL schema of 22 domain tables and 27 foreign keys. It ships a 3D body map sculpted in code as a signed distance field, rule-based personal offers priced by one function for the bag and the charge, a row-locked idempotent checkout, card payments confirmed only by signed webhooks, drug-interaction warnings, refill reminders, staff roles with authenticator-app two-step sign-in, 24-hour prescription auto-accept, three-level bans that see through Gmail aliases, server-side rendering, a nonce CSP, ClamAV-scanned uploads, and 155 PHPUnit plus 56 Playwright tests, with Lighthouse 100 for accessibility, best practices and SEO. Role: Solo build — research, UI/UX, database, backend, admin panel, security, SEO, tests and docs. Stack: Laravel 12, PHP 8.2, React 19, Inertia 2 + SSR, Tailwind CSS 4, MySQL, Redis, Stripe, GSAP, Three.js, Leaflet · OSM, Resend, ClamAV, Playwright. Security baseline: Row locking, Idempotent checkout, Offers redeemed under lock, Signed payment webhooks, Staff roles + TOTP, ClamAV upload scanning, Nonce CSP, Named rate-limit counters, RequestGuard, reCAPTCHA v3 + v2, Enumeration-safe responses, Private prescription disk, Three-level bans, Admin 404 + idle timeout. Metrics: 1,186 real products · 22 database tables · 211 automated tests · 100 Lighthouse a11y · SEO. Highlights: A store that notices: every visitor (guest or customer) builds a quiet profile of views, visible time, bag adds and purchases, and six explainable rules turn it into time-limited personal offers, capped at 15% and redeemed exactly once inside the checkout transaction. Stock that can't be oversold and a price that can't drift: products locked in id order, a conditional decrement that refuses to go below zero, one-time checkout tokens, and the same Pricing function used for the bag preview and the order. Medicine safety built in: the bag checks every product's active ingredients against 24 well-established interaction rules (two paracetamol products, blood thinner + NSAID, sildenafil + nitrate…), serious ones must be acknowledged and are shown to the pharmacist, and regular buyers get a refill reminder before they run out. Built for pharmacy staff, not developers: a 404-for-everyone-else admin panel with "Today at Zovita" to-do cards, plain-language help on every heading, prescription review with 24 h auto-accept, customer timelines and a ban wizard — all usable on a 320 px phone. How it works: Checkout runs in one transaction (retried up to three times on deadlock): products are read with SELECT … FOR UPDATE in ascending id order, then stock is decremented with UPDATE … WHERE stock >= qty, so it can never go negative even if a lock were bypassed. orders.checkout_token and orders.number are both unique: a double-submitted checkout returns the order already placed, and a random order-number collision retries inside a savepoint without aborting the transaction. Personal offers come from OfferEngine's six rules (hesitation 8%, cart rescue 5%, regular 10%, welcome 5%, loyalty 5–10%, comeback 7%); Pricing::apply() prices both the bag and the order, caps the total at 15%, and checkout locks the visitor's live offers FOR UPDATE before marking them redeemed with the order id. Personalisation uses one visitor key — u:{id} when signed in, g:{uuid} from a first-party HTTP-only cookie otherwise — and signing in merges the guest's interaction counters and live offers into the account in one locked transaction. Every route throttle names its own counter (throttle:6,1,checkout.store): Laravel's default key is per user or IP, so bag and wishlist activity used to eat into checkout's limit. A test proves it can't any more. RequestGuard rejects odd methods, URLs over 2,048 characters, more than 40 query parameters, oversized bodies and NUL bytes before any session or database work; a per-client budget and nginx limit_req sit on top. Bans have three levels; the deep level blocks the canonical email (Gmail dots and +aliases folded), phone, device cookie, browser fingerprint, IP and its /24 or /64 network across the whole site. The 3D body map is a signed distance field meshed with surface nets by a Node script and baked into a 463 KB Brotli binary where every vertex carries its body-region id for picking and glow. Card payments use Stripe Checkout (a built-in sandbox gateway locally): the order is created awaiting payment with stock reserved for 30 minutes, and only a webhook with a valid HMAC-SHA256 signature, a timestamp under five minutes old, an event id never seen before and the exact amount and currency marks it paid; payments:expire cancels and restocks unpaid orders and refunds money that arrives late. InteractionChecker maps each product's stored generics to drug classes with tolerant patterns (so the catalogue's own spellings still match) and ignores topical forms, so a ketoconazole shampoo never warns about a statin. Refill reminders take the median gap between a customer's real orders of the same medicine (7–120 days) and e-mail three days before it should run out, once per purchase cycle (a unique key on the last purchase date), with a signed one-tap reorder link. Staff sign in with a password, and once they enable an authenticator app every sign-in also needs a TOTP code; used time-steps are stored so a code can't be replayed, recovery codes are hashed and single-use, and permissions per role (owner, pharmacist, support) are checked on every admin route. The Urdu interface is a 1,460-entry dictionary applied to the rendered DOM; headline line-splitting is skipped in Urdu because it ran before translation and handed the dictionary sentence fragments, and every font stack falls back to Nastaliq after monospace fallbacks drew the letters unjoined. A scroll reveal left the "Shop by concern" grid invisible on every screen: the cards had a CSS transition on opacity, so when ScrollTrigger refreshed, gsap.from() re-read their mid-transition opacity (about 0) as the target. Explicit fromTo() end values fixed it. A responsive sweep of every storefront, account and admin page at 320–1920 px found the admin dashboard 698 px wide on a 375 px phone (a grid with no column template); grid-cols-1 fixed it and a Playwright test now guards the admin pages. Links: Source on GitHub: https://github.com/ahmershahdev/zovita; Case study · Building Zovita+: https://ahmershah.dev/blogs/zovita-case-study-online-pharmacy-platform; Article · The discount that knew you were hesitating: https://ahmershah.dev/blogs/the-discount-that-knew-you-were-hesitating-rule-based-personalisation-in-zovita ### BookMyMovie — Cinema Ticketing (2026) Pick your seat, watch the trailer, and never share it with a stranger. Think about the last time you bought a cinema ticket in Pakistan. Maybe you phoned the cinema, maybe you queued at the counter, or maybe a website let you pick a seat and then told you at the very last step that someone else already had it. BookMyMovie is my fix for that night out. You watch the trailer, choose your exact seat on a live map (or sit in it first, inside a 3D model of the hall), add popcorn, and walk in with a QR ticket on your phone that still opens when the lobby has no signal. Behind the counter, cinema staff get their own back office to scan tickets, refund bookings and see how tonight is selling. It began in July 2026 as my Aptech DISM project and grew into a complete cinema platform that I built alone, from the first sketch to the database, payments, security and the help pages. The rule I set myself on day one was simple: never sell the same seat twice, and never show a price you won't charge. Everything else came from imagining real evenings. Friends who want to split the bill, a sold-out show where you'd happily wait for a cancellation, a parent mixing adult and child tickets, someone who'd rather read the site in Urdu. Each of those became a feature, and each one had to keep that first rule intact. Role: Solo build — research, UI/UX, database, backend, payments, security, SEO and docs. Stack: Laravel 12, PHP 8.2, React 19, Inertia 2, TypeScript, Tailwind CSS 4, MySQL, Three.js, Motion, Service worker, JazzCash. Security baseline: Row locking, Status compare-and-set, Idempotent checkout, IDOR scoping, HMAC-signed tickets, Nonce CSP, reCAPTCHA v3 + step-up, Argon2id hashing, Admin TOTP, Account/IP/device bans, Audit log. Metrics: 56 database tables · 49 screens · 0 seats sold twice · 100 Lighthouse a11y · SEO. Highlights: Two people can press Confirm on the same seat in the same half-second and only one walks away with it. The other is told straight away, instead of finding out in the lobby. In a test with eight buyers racing for one seat, exactly one got it. Made for a phone in a dark cinema lobby: tickets that open offline, Apple and Google Wallet passes, trailers that only play on a fast connection, and no page that ever scrolls sideways on a small screen. A back office for the people who actually run a cinema: an owner, managers and box-office staff each see only what their job needs, sign in with a code from their phone, scan tickets at the door and refund a booking with one button. How it works: Holding seats and checking out lock rows with SELECT … FOR UPDATE in one order everywhere (customer → show → seats → cart) inside a transaction that retries deadlocks up to three times. booking_seats has UNIQUE (show_id, seat_id, seat_lock); seat_lock is 1 for a live ticket and NULL once cancelled, so a cancelled seat can be resold but the database refuses a second live ticket for the same seat. Checkout carries an idempotency key under a unique index, so a double-tapped Confirm returns the booking already made; booking status moves through Booking::claimStatus(), a guarded UPDATE that fails if the status changed, so a booking can't be cancelled or refunded twice. The waitlist job is ShouldBeUnique per show and uses WithoutOverlapping; home-page caches use Cache::flexible so one locked request rebuilds a stale value instead of a stampede. Every customer lookup is scoped to the signed-in user and returns 404 for someone else's booking; share links use 40-character random tokens; ticket QR codes carry an HMAC-SHA256 signature checked with hash_equals at the door; admin routes fail closed to the highest role. Bots meet a honeypot, a fill-time check, a server-drawn single-use image code and invisible reCAPTCHA v3, with the v2 checkbox only as a step-up when v3's score is low. SEO without SSR: server-rendered titles, canonical links, per-film Open Graph images, JSON-LD Movie and ScreeningEvent data, a generated sitemap and llms.txt / llms-full.txt for AI assistants. A production build once merged the seat-map page into a shared chunk, dropping it from Vite's manifest and turning the page into a 500; a test now fails if any page is missing from the manifest. Links: Source on GitHub: https://github.com/ahmershahdev/bookmymovie; Case study · Building BookMyMovie: https://ahmershah.dev/blogs/bookmymovie-case-study-cinema-ticketing-platform; Article · One seat, two buyers: https://ahmershah.dev/blogs/one-seat-two-buyers-race-conditions-idor-and-keys-in-bookmymovie ### GleanGrid — Farmers Market Pre-Orders (2026) Reserve this week’s harvest online, pay the farmer at the stall. GleanGrid puts Hyderabad’s farmers markets online without taking the market away. Farmers publish weekly stock, prices and pickup windows; customers find a market on a live map, reserve produce against real stock, pay online or at the stall, and collect it. Admins approve stalls, moderate listings, refund payments and read platform-wide reports. Built solo for Aptech TechWiz 7: Laravel 12 and React 19 joined by Inertia with server-side rendering, on a MySQL schema with foreign keys, 22 CHECK constraints, reporting views and triggers. Three roles, 62 screens, eight languages including right-to-left Urdu and Arabic, JazzCash / Easypaisa / card payments with automatic refunds, Google and Facebook sign-in, and a checkout stress-tested with fifty buyers racing for the last item at the same instant. Role: Solo build — research, UI/UX, database, backend, payments, security and docs. Stack: Laravel 12, PHP 8.2, React 19, Inertia.js SSR, Tailwind CSS 4, MySQL, JazzCash, Leaflet · OSM, Motion, Vite, Playwright. Security baseline: Per-response CSP nonce, Per-load CSRF tokens, Signed form tickets, Turnstile + reCAPTCHA v3, Argon2id hashing, Rate limiting, Row locking, Idempotent payments, HMAC callbacks, E-mail OTP. Metrics: 62 screens · 50 buyers, one sale · 133 automated tests · 8 languages. Highlights: A pre-order flow that never oversells: stock, coupons and pickup-window capacity are all decided under row locks inside one transaction — proven with 50 real parallel processes, not just unit tests. Payments that can’t double-charge: a locked state machine, one-time idempotency keys, amounts recomputed on the server, HMAC-verified gateway callbacks, and automatic refunds when a confirmation arrives after the payment window closed. Three complete workspaces (customers with QR pickup passes, price history and restock alerts; farmers with weekly stock and a scanner; admins with approvals, badges and a payments console), built for real phones: bottom-sheet filters, a live map, 8 languages with RTL and Lighthouse 100 for accessibility, best practices and SEO. How it works: Overselling is prevented with SELECT … FOR UPDATE row locks on the products, taken in id order (so two checkouts cannot deadlock) inside one database transaction; duplicate basket lines are merged before the stock check. Coupons and the pickup-window row are locked the same way. The pickup-window booking count is itself a locking read, because under MySQL's REPEATABLE READ a plain COUNT after waiting for a lock reads a stale snapshot: a 50-process stress test first let 20 orders into a 3-place window, and that one-line change fixed it. A per-customer atomic cache lock stops double-submitted checkouts, deadlocks retry up to three times, and a CHECK constraint keeps coupon use under its limit even if a lock is ever forgotten. Quieter races: the farmer's stock form sends back the number it showed and the server subtracts sales made meanwhile; the weekly restock is one atomic UPDATE; restock alerts are claimed with a conditional UPDATE so each is sent exactly once. Payments: the payment row is locked and moves pending → processing once, every attempt carries a one-time idempotency key under a unique index, amounts are recomputed on the server, JazzCash callbacks are HMAC-SHA256 verified, confirmations after the 20-minute window are refunded automatically, and card numbers are never stored. Bot and CSRF defence: a honeypot, a signed and encrypted form ticket issued with every page, invisible reCAPTCHA v3 plus a Cloudflare Turnstile or reCAPTCHA v2 challenge, CSRF tokens regenerated on every page load, cross-site writes refused, and a new CSP nonce on every response. Tested with 133 PHPUnit tests (779 assertions) on a real MySQL database, Playwright end-to-end tests on desktop and phone, and GitHub Actions on PHP 8.2 and 8.3. Links: Source on GitHub: https://github.com/ahmershahdev/GleanGrid; Case study · Building GleanGrid solo: https://ahmershah.dev/blogs/gleangrid-case-study-farmers-market-preorder-platform; Article · The last mango problem: https://ahmershah.dev/blogs/the-last-mango-problem-race-conditions-in-a-real-marketplace ### Commerza — E-Commerce Store (2026) A security-first e-commerce engine, built without a framework. Commerza is an online watch store I built from scratch in plain PHP and MySQL. Shoppers browse collections, compare watches, keep a wishlist and check out with cash on delivery or a Stripe card, then track the order to their door. Behind it sits an admin panel where the owner and sub-admins run products, orders, refunds, coupons, reviews, site content and security, each seeing only the tabs their role allows. I built the first version in three months without Laravel or React, to learn what a framework normally does for you. Version 2.0.0 is the grown-up release. Pages now live in one folder with a shared header and footer instead of seventeen copies. The database seed is cleaned of anything private. Every push runs a Docker stack through a security probe, an SEO audit, a 15-user load test and Playwright tests on desktop and mobile. The repository is now archived as a finished, MIT-licensed reference project. Role: Solo build — architecture, backend, storefront, admin panel, security, testing and docs. Stack: PHP 8, MySQL / MariaDB, Apache, JavaScript, jQuery, Bootstrap 5, Stripe, Cloudinary, Docker, GitHub Actions, Playwright. Security baseline: Nonce-only CSP, CSRF tokens, reCAPTCHA v3 + v2 + fallback, Rate limiting, Argon2id hashing, Admin email 2FA, Stock row-locking, Upload malware scanning, Audit logs, Customer blacklists. Metrics: 346 tracked files · 143 commits · 0% errors, 15 users · 33/33 security checks. Highlights: Fifteen simulated shoppers on fifteen different journeys at once (browsing, searching, filling carts, signing up) made 3,243 requests with zero errors, and 95% of pages answered within 637 ms. Two checkouts can never sell the same last watch: stock rows are locked inside a transaction, every order carries an idempotency key, and high-value cash orders need an emailed one-time code. A full back office: products with a trash bin, orders with delivery dates and refunds, coupons, review moderation, slider and SEO settings, and sub-admins whose access can be suspended instantly. How it works: Plain PHP 8.1+ on Apache: clean URLs rewrite to one controller per page in pages/, shared markup renders through templates/partials, and every request boots through a single backend/core/data.php. 43 InnoDB tables and 11 reporting views; every request-derived value goes through a prepared statement, IDs are cast to int and enum inputs are checked against allowlists. Checkout and refunds run in transactions with SELECT … FOR UPDATE, conditional stock decrements and idempotency keys; wishlist creation is an ON DUPLICATE KEY upsert. The storefront Content-Security-Policy allows only scripts carrying a fresh per-request nonce; inline onload/onerror handlers were replaced by one nonce-tagged loader. Uploads are checked by type, scanned for PHP tags, eval,