Our work

What we actually built, and what it took.

Real projects, described without naming the client. No logos, no testimonials — just the problem, what we did about it, and what changed.

Real estate · UAE

A property search bot brought back from a dead API

The problem

A property firm ran a Telegram bot that answered unit and availability questions for their sales team. It had stopped returning results, but nobody noticed for days because the bot still replied — it just replied that it could not find anything. Over ninety per cent of searches were quietly failing, and the sales team had started assuming the inventory was empty.

What we did

  • Pulled the bot's own logs apart to separate real 'no match' answers from silent failures, which is what revealed the true failure rate
  • Traced the failures to an upstream data provider whose access had lapsed weeks earlier, with no alert configured anywhere
  • Evaluated replacement data sources against the queries the bot was actually being asked, not against a feature list
  • Migrated the integration to a working provider and rewrote the failure path so an upstream outage now surfaces as an explicit error instead of an empty result
  • Took a full filesystem backup of the server before touching anything, so the old behaviour stayed recoverable

Outcome

Searches returned real results again the same day, and a second pass cut the average lookup from roughly 31 seconds to under 4 by removing two redundant browser scrapes per request. More usefully, the bot can no longer fail quietly.

telegram botsapi developmentmaintenance support

Email marketing · India

A million-record contact list, validated before a rupee was spent

The problem

A large contact list was about to be pushed through a paid verification service. Verification is billed per address, so anything unusable in the list would be paid for twice — once to verify it and again in damaged sending reputation if it slipped through.

What we did

  • Profiled the file first — encoding, structure, duplicates — before assuming anything about its contents
  • Removed exact duplicates, then institutional addresses that should never receive commercial mail
  • Ran an MX lookup across every unique domain in the list, so any address whose domain cannot receive mail at all was removed before the paid step
  • Checked a control sample by hand, which corrected an early assumption that certain legacy consumer domains were dead — their mail servers are still live
  • Repartitioned the cleaned data into batches sized to the verification service's actual rate limit, so the run time could be predicted rather than guessed

Outcome

Roughly one in eight records was removed before paid verification ran, across 284,000 unique domains. The DNS sweep cost nothing but the time to run it.

email infrastructureweb scrapingpython development

Marketing services · India

A sending platform moved between servers without losing a campaign

The problem

A live email marketing platform had to move to a new server. It was in daily use, so the migration could not take the system offline for long, and the sending reputation attached to its domains had to survive the move intact.

What we did

  • Took a complete backup before the source server was touched — and dumped the databases separately, because a file-level copy does not capture a running database cleanly
  • Rebuilt the application stack on the new host and restored data into it, then fixed the database role grants that a restore does not carry over
  • Reconfigured authentication records for every sending domain so mail kept passing SPF, DKIM and DMARC from the new IP
  • Verified delivery end to end against an external authentication checker before any production traffic moved across
  • Kept the old server intact until the new one had been proven, rather than decommissioning on the same day

Outcome

The platform came up on the new infrastructure with its data and sending configuration intact. The separate database dump turned out to be the step that made the restore possible at all.

cloud serveremail infrastructuredevops

Real estate · UAE

A manual research job turned into an internal web tool

The problem

A team was compiling property and ownership information by hand, one record at a time, from a public portal. It was slow, the output format varied between people, and there was no record of what had already been pulled.

What we did

  • Watched how the work was actually being done before designing anything, which showed the real bottleneck was re-checking earlier results, not the lookups themselves
  • Built an internal web tool that takes a link, confirms what it found before running, and produces a structured spreadsheet export
  • Added a history view so the team can see what has already been extracted instead of repeating it
  • Tested against a known-good control case first — which surfaced that one whole category of property returns no data at all, a limit of the source, not a bug
  • Deployed it behind access control on the client's own server, under a service manager so it survives a reboot

Outcome

Work that took hours now takes minutes, and the output is consistent regardless of who runs it. The control test also set honest expectations about what the data source can and cannot return.

web scrapingcustom softwareweb development

More of what we have shipped

Shorter, but no less real — the same rule applies: no client names, and no figure that was not measured.

  • Media · India

    Own mail server instead of per-mailbox fees

    Stood up a self-hosted mail platform on the client's own VPS, with DKIM signing, SPF, a published DMARC policy and reverse DNS matching the sending hostname.

    Outbound mail passes SPF, DKIM and DMARC at Gmail. Mailboxes stopped being a per-user monthly cost, and a second domain now runs on the same server.

    email infrastructurecloud server

  • Furniture retail · UAE

    A client given their own sending account, not a shared one

    Separated one client onto their own customer account inside a shared email platform, with four dedicated sending servers, their own authenticated domains and their own timezone.

    They see only their own campaigns and servers. Their sending reputation is theirs alone — another tenant's mistake cannot damage it.

    email infrastructurecloud server

  • Marketing services · India

    A platform rescued from an expiring domain

    A sending platform was tied to a domain about to lapse. Moved the application to a new domain, rewrote the URLs held inside its own database, reissued SSL and re-pointed DNS.

    The platform kept running through the domain change. The URLs stored in the database were the part that would have broken it silently.

    cloud serveremail infrastructuredevops

  • Media · India

    A fresh sending platform, deliberately wiped to zero

    Cloned a working install onto a new server, then truncated 130 tables to remove every subscriber, campaign, log and delivery server the source carried.

    A genuinely clean instance rather than someone else's data with a new logo on it. Lookup and system tables were kept so nothing had to be rebuilt by hand.

    cloud serveremail infrastructure

  • Internal tooling

    An email verification tool built around a real rate limit

    Built a verification service with a worker and a web UI, sized to the provider's actual request ceiling rather than an optimistic guess.

    Run times became predictable instead of a surprise. Setting the rate limit correctly before a large list is the difference between an accurate estimate and a meaningless one.

    email infrastructurepython developmentcustom software

  • Email marketing · India

    A four-minute manual job reduced to a fraction of a second

    Replaced a manual spreadsheet process with an indexed pipeline that takes a verifier's output and produces an import-ready file.

    About four minutes per batch became 0.15 seconds. The old version re-read 66 MB of spreadsheets on every run; the new one does not.

    python developmentemail infrastructure

  • Email marketing · India

    “Zero bounces” that actually meant nothing was listening

    A campaign platform reported no bounces at all. Traced it to feedback notifications never being wired up, then connected the provider's bounce and complaint stream properly.

    Bounces and complaints started being recorded. A zero bounce rate is almost never good news — it usually means nobody is counting.

    email infrastructureapi development

  • Media · India

    An AI-assisted news publication, built end to end

    Designed and built a multi-category news site with an editorial pipeline, an admin area and its own mail server, running on the client's infrastructure.

    Live and publishing, with the newsletter as the revenue asset and the site as its growth engine.

    web developmentai automationcloud server

  • Clothing retail · India

    A shop management portal for a two-partner business

    Built a stock, order and billing portal with role-based logins, backed by Postgres, and deployed it under a process manager with automatic SSL renewal.

    Both partners work from the same numbers instead of separate spreadsheets. A regional-language Telegram bot was added alongside it.

    custom softwareweb developmenttelegram bots

  • Home repairs · Georgia

    A local repair business taken online

    Took a design to production on a small cloud server — real build, service manager, reverse proxy and SSL, sized to a 1 GB instance.

    Live on the client's own infrastructure. A backend and form handling were deliberately left out because the business did not need them yet.

    web developmentcloud server

  • Software services · India

    A client portal on self-hosted auth, not a subscription

    Built a quote-and-payment tracking portal on a self-hosted authentication and API layer over Postgres, with row-level policies so a signed-in user sees only their own records.

    Signup, login, role checks and the contact form verified end to end. No per-user platform fee.

    custom softwareweb developmentcloud server

  • Marketing · India

    A retired site brought back and re-pointed in an afternoon

    A marketing site had been shut down. Rebuilt it on current infrastructure, created every DNS record by hand after the provider's import pulled in none, and issued SSL.

    Back online the same day. The empty DNS import is a trap worth knowing about — it reports success and creates nothing.

    web developmentcloud server

  • Local business · India

    Small business sites that stay up without a retainer

    Built and deployed several local business sites — a beauty studio, a computer training institute, a personal portfolio — as static builds behind a shared reverse proxy with automatic certificates.

    Each costs effectively nothing to keep online and needs no monthly platform. Certificate renewal is automatic.

    web developmentcloud server

  • Trading · India

    A market alert bot that deliberately cannot trade

    Built an index signal bot that reads live market data and sends alerts. It has no order-placing code path at all — not disabled, absent.

    Alerts without the risk of an automated mistake. Rate limiting had to be built around response size rather than call spacing, which is what the provider actually enforces.

    telegram botspython developmentapi development

  • Own product

    WhatsApp Business API wired into an admin panel

    Integrated the official WhatsApp Cloud API: signed webhook verification, a structured qualification flow, opt-out recorded in the database, duplicate-delivery guards, and a message inbox in the admin panel.

    Running on this site. Opt-outs are a stored record rather than a promise, which is what a platform review asks to see.

    whatsapp automationapi developmentcustom software

  • Own product

    A find-audit-rank pipeline for outreach

    Built a panel that finds local businesses, audits what they have online, scores who is losing the most, and drafts a first message. Sending is one press per lead, never a bulk blast.

    A prospect list becomes a ranked shortlist with a reason attached. An estimated-money-lost figure was built, then removed — it rested on invented averages, and an invented number costs more credibility than it buys attention.

    web scrapingai automationcustom software

  • Email marketing · India

    Campaign templates generated from a chat box

    Wired a language model into an internal chat so campaign templates could be produced from a brief, then pushed into the sending platform.

    Template production stopped being a manual HTML job. Writing templates back through the wrong tool corrupts them — the pipeline uses a method that does not.

    ai automationemail infrastructure

  • Infrastructure

    Eight sites consolidated onto one server

    Moved eight separate sites and applications onto a single VPS behind one reverse proxy, each with its own certificate, all services set to restart on boot.

    A reboot brings everything back with no manual steps — verified by rebooting it. One bill instead of several.

    cloud serverdevops

  • Infrastructure

    DNS moved between providers without dropping mail

    Migrated multiple domains onto a managed DNS provider, rebuilding records by hand where the automatic import brought across nothing, and keeping mail records pointed at the live mail host throughout.

    No mail interruption during any move. The risk in a DNS migration is almost always the MX record, not the website.

    cloud serveremail infrastructure

  • Software services · India

    An auth service that crash-looped after a restore

    A self-hosted authentication service would not start after a database restore. It was a permissions problem, not a corruption one — role search paths and schema ownership, not the data.

    Back up without restoring from an older backup. Database grants do not survive a restore, and the symptom looks nothing like the cause.

    cloud serverdevopscyber security

  • Internal tooling

    A remote server tool behind proper authentication

    Built a remote tooling service reachable over the internet, protected with an OAuth flow and PKCE rather than a shared token in a URL.

    Usable from anywhere without leaving an open door. Running as a managed service, it comes back on its own after a reboot.

    api developmentcyber securitycloud server

  • Infrastructure

    A server that looked dead and was serving fine

    A site appeared completely down from the office. It was a routing problem on one carrier's path, not an outage — proven by reaching the same server over IPv6 and from outside the network.

    Fixed by putting the domain behind a proxy network instead of rebuilding anything. Diagnosing from a single connection is how a healthy server gets wrongly declared dead.

    cloud servermaintenance support

Client names, logos and quotes are deliberately absent from this page. Work done under a confidentiality arrangement stays confidential, and we would rather describe the engineering honestly than decorate it with borrowed credibility.

Let's talk

Got a problem shaped like one of these?

Describe what is going wrong. We will tell you what it actually takes to fix, and what it does not.

Get a free consultationWhatsApp us