Amul R Build Log

I build the whole thing, and I check that it actually works.

Review received

“Excellent work! Mr. Amul upgraded my MetaTrader 5 MQ5 code in just 2 hours, and the result was fully satisfactory and exactly as I assumed. As I'm new on the platform, his speed and professionalism helped me a great deal. Highly recommended!”

Client · Freelancer.com · 11 Sep 2026 · 5.0 / 5

10× faster Delivered — as fast as a few hours

5× less In cost — as low as ₹999

Against the usual agency quote for the same brief. The job reviewed here: 2 hours, ₹999.

Support after delivery — I stay on after handover. Anything that breaks in the first 30 days, I fix at no charge.
Quality you can check — every build is rendered and tested on a real phone before I call it done, and every figure on this page was read off the running system.

Storefronts, web products, dashboards, trading tools, content pipelines and the automation between them — designed, built and shipped solo. Fixed price, fixed delivery date, both agreed before I start.

Selected work

Shopify WhatsApp ordering Built to demonstrate

Nila Silks

A Shopify storefront for a South Indian textile shop, built the way a local shop actually sells: browse online, tap Order on WhatsApp on any product and the message arrives with the item, price and link already filled in. Twelve products across men, women and kids, Tamil name lines, UPI-or-cash and collect-at-the-counter written into the page.

Full record

The WhatsApp pieces are theme code, not an app subscription: a block after the buy buttons, a floating button on every page, and the shop's number as a single theme setting the owner can change without touching code. Collections are tag-driven, so the Diwali shelf fills itself when stock is tagged.

What is placeholder, said plainly

The shop name, number and product photography are stand-ins for the demo — the real shop's photos and details go in on launch. The structure, the ordering flow and the checkout are the deliverable.

A practice question rendered in Tamil with four answer options, above a progress and streak bar. — A practice question in Tamil. The whole product runs in both languages.

My own product In production Bilingual

Brain Grain

A free exam-preparation platform for Tamil Nadu, live and in daily use. It is the reference for what I can run end to end: accounts, payments through Razorpay, a Postgres database, event analytics, a generated content pipeline, and a Kotlin Android client that works offline.

Full record

The content is not a CMS full of typing. Question banks, concept pages and exam packs are generated and validated by scripts, then published — which is why five exam tracks and several thousand questions are maintainable by one person, in English and Tamil.

Why this matters for your project

Every hard part of a commercial web build — sign-in, taking money, a schema that migrates, knowing whether anyone actually used the feature — is something I run in production today rather than something I have read about.

The Brain Grain homepage showing live usage tiles for learners, points, minutes and questions answered, beside a top-learners leaderboard. — Homepage — usage tiles and the leaderboard read from live analytics and the database.

Two phone screens side by side. Left: a timed Blitz round with a history question and four options above a Skip button. Right: a full-screen Tamil concept reel on a dark starfield with a progress bar and side action buttons. — The two screens the mobile app is built around — Learn, a timed question loop, and Watch, a swipeable concept reel in Tamil. Captured at phone size on the live site; the Android app renders the same surfaces.

My own product MQL5 On sale

Risk Guard

A risk-enforcement tool for traders on funded and prop-firm accounts, where one bad day ends the account. It never opens a trade; it watches every position and enforces the rules the trader sets — daily and weekly loss locks, trailing drawdown, lot caps, forced stop-loss, a trading window. Free version on the MQL5 CodeBase; the prop-firm build is on the MQL5 Market at US$39.

Full record

The same rule set exists three ways: the MetaTrader 5 build, a cTrader port in C#, and a TradingView version in Pine Script, open source on GitHub. Every rule is an input and can be switched off; locks are stored in terminal variables so a restart does not clear them.

What the marketplace validator caught

Three rejections before it passed, each a real bug: the validator tests netting accounts, so an opposite trade shrank the position and its stop could not attach; its huge test deposit made 1% risk exceed the symbol's lot limit; and on two-digit gold a 20-pip stop implied a size no margin could cover. All three are handled in the code now, and a real broker would have thrown each of them too.

Verification is scripted, not eyeballed: a self-test expert runs in the Strategy Tester headless and asserts that every lock fires.

Shopify Liquid Built to demonstrate

Rootwood Furniture

A complete solid-wood furniture storefront on Shopify's Horizon theme — twenty products across five automated collections, an eight-field specification schema carried on every product, five custom theme sections written in Liquid, and five policy pages.

Full record

The specification table is the part worth looking at. Material, finish, dimensions, warranty, assembly and care are structured product data, not a table typed into a description — so they render identically on every product, stay fillable by whoever adds stock, and cannot be formatted three different ways by three different people.

What rendering caught that reading the code did not

A room card silently vanished from a grid whose own subheading promised five rooms: the collection handle was study, the real one was study-office. No error, no warning, no failed build — just four cards where there should have been five.

The specification table rendered empty on all twenty products. Its eight field definitions had imported and were readable by the storefront; its 160 values had never imported at all. The admin looked correct and the store showed nothing.

A product specification table listing material, finish, width, depth, height, assembly and warranty, with a care note beneath it. — The specification table, rendered from structured product data — not typed into a description.

A school admissions page headed Admissions open for 2026-27, with figures for years, students, ratio and pass rate, and a grid of quick actions. — The rebuilt school page, generated by the pipeline. The banner at the top marks it as a sample and the school as fictional.

Pipeline Multi-tenant tool Service pilot

School site & admissions desk

A pipeline that reads a school's existing public website and rebuilds it as a fast, mobile-first page, plus the admissions dashboard the enquiries land in. Both halves are live and clickable.

Full record

The two are wired together, which is the part most demos skip: submit the enquiry form on the sample school site, then open the dashboard and your own enquiry is sitting at the top of the pipeline. Each school gets a single access code rather than user accounts, and every query is scoped to the school that code resolves to.

Where naive scraping breaks on real sites

School phone numbers are landlines, and a mobile-format check rejected three real schools as having no contact number at all. Campus photographs live in lazy-loading attributes, so reading the obvious one returned clipart and missed every real photo of the building.

Two EURUSD five-minute charts stacked. Before: thin trendlines project forward from each pivot only. After: thick green and red connector lines join consecutive pivot lows and highs, with the original trendlines still drawn on top. — Before and after on the same EURUSD M5 window: the client's indicator only projected forward from each pivot; the delivered build joins consecutive pivots — lower on higher lows, upper on lower highs — with the original lines kept on top.

MQL5 Client work Freelancer.com

MetaTrader 5 indicator upgrade

A trader's custom MQL5 indicator drew trendlines forward from each pivot but never the stretch between two pivots. He sent the source and a screenshot with the lines he wanted drawn by hand. Delivered the same day: two new connector plots, lower for a higher low and upper for a lower high, straight on lower timeframes rather than stepped, with toggle, colour and width inputs and every change commented in the source.

Full record

Two follow-ups on the same day, both at no charge because they were in the spirit of the brief: the vertical joins where a line jumps to a new pivot, and a trendline that had become invisible because the new thicker plot was painting over it. That second one was found by comparing every buffer of the old and new builds value for value rather than by eye.

What the client wrote

"Excellent work! Mr. Amul upgraded my MetaTrader 5 MQ5 code in just 2 hours, and the result was fully satisfactory and exactly as I assumed. As I'm new on the platform, his speed and professionalism helped me a great deal. Highly recommended!" — client, via Freelancer.com, 11 Sep 2026, 5.0

The RailCall marketplace listing for the Razorpay module: Finance category, v1.0.0, publisher-signed badge, an overview of the governed commands, and the one-line CLI install command. — The public listing on railcall.ai: publisher-signed, free, installed with one CLI command. Submitted as the Round 2 contest entry; in the marketplace review queue at the time of writing.

Python Payments integration Contest entry

Razorpay module for RailCall

RailCall is a governance layer for AI agents: every command that touches an outside system goes through an “airlock” where a human approves the exact payload before it is sent. Their developer challenge asked for marketplace modules; I built the Razorpay one in a day because I run Razorpay for real and nobody else was going to build it. Eleven commands: payments, payment links, orders, settlements and subscriptions as reads; refunds, link creation and cancellations gated behind approval.

Full record

The refund is the point. Real money leaves the merchant account, so the preview shows what was captured, what is already refunded and what this one approval will send back. Credentials sit in the station vault and are read only at handler time — never an input, never stored by the module. Proven end to end on a Razorpay test account: a ₹499 payment, a ₹100 refund approved through the airlock, and the same approval replayed returning the existing refund rather than a second one.

The catch

Razorpay has no idempotency header on refunds, so an agent that retries a timed-out call refunds twice. The module writes the approval's payload hash into the refund's notes and checks existing refunds for that hash before creating one. The replay test above is what proves it.

Their station also silently redacts any output field with “key” in its name to •••• — which hid the idempotency value in every receipt until it was renamed. Found by reading the receipt, not the docs.

Tell me what is broken, or what you want built.

Useful first message: the site or store, what should happen that currently does not, and your deadline. I will reply with what I think is actually wrong — before any quote.

I build the whole thing, and I check that it actually works.

What I take on

Shopify

Web products

Dashboards & tools

Pipelines & automation

Selected work

Nila Silks

Brain Grain

Risk Guard

Rootwood Furniture

School site & admissions desk

MetaTrader 5 indicator upgrade

Razorpay module for RailCall

How I work

I look at the page before I call it done

I check the copy against the data

Anything that writes has a dry run

I go looking for the silent failure

Prices & delivery times

Rescue

Full site

Shopify store

Custom build

Book it

Tell me what is broken, or what you want built.