Product engineering studio · Kolkata

We build AI platforms that turn more of your leads into customers.

We are a product engineering studio. We design and engineer custom AI systems — lead qualification, sales and ops automation, and the analytics layer that proves it worked.

11 platforms shipped for teams in real estate, fintech and healthcare

Shipping for
Asobr Nova Wealth K-BID Bensen Meridian Health MyPrime Network Larkfield Legal Orbit Logistics Cedar & Co Realty Vantage Dental Group
The gap

Leads go cold before you follow up. Good prospects get lost in the noise.

Your team answers the easy enquiries first and the serious buyer waits until tomorrow. Nobody knows which channel actually produced revenue. More headcount doesn't fix a routing problem.

Qualify in seconds, not days

An AI layer reads every inbound message, asks the qualifying questions your best rep would, and scores intent before a human opens the tab.

Route to the right human

Hot leads land in the right inbox with context attached. Everything else is nurtured automatically until it's worth a call.

Prove what worked

One dashboard tying source to qualified lead to closed deal — so budget goes where the revenue actually came from.

Full disclosure

What we won't sell you

You have almost certainly been pitched at least three of these. We have sat through the same decks.

  • A chatbot that reads your FAQ page back to customers
  • A 40-slide "AI transformation roadmap" with no line of code in it
  • A six-month pilot that quietly never reaches production
  • "Synergistic, next-generation, AI-first digital solutions"
  • What you get instead: a working system in production, and the number it moved.
What we build

Seven things, done properly

All services →

AI Lead Qualification

Every enquiry answered in under a minute, scored and routed with full context.

Dedicated Hiring

Vetted developers working only on your product, on a monthly contract — interviewed by you, replaced by us.

Sales & Ops Automation

CRM hygiene, follow-up sequences and handoffs that run without anyone remembering to.

Custom AI Platforms

Retrieval, agents and evaluation harnesses built on your data — owned by you, not rented.

Analytics Dashboards

Source → qualified → closed, in one view your leadership team actually opens.

Product Engineering

Web and mobile products shipped by a small senior team, from spec to production.

Performance Marketing

Paid acquisition run against qualified-lead cost, not clicks — feeding the same AI that qualifies it.

Not sure which of these you need? That's what the fit call is for.

Book a call →

One closed loop, not two vendors

We buy the traffic and own the system that qualifies it, so spend is judged on booked meetings.

01 · SPEND

Campaigns built on intent

Google, Meta and LinkedIn campaigns targeted with the segments your closed deals actually came from.

02 · QUALIFY

Every click meets the AI

Landing pages hand straight to the assistant, so paid leads are scored in seconds instead of sitting in a form queue.

03 · ATTRIBUTE

Optimised on revenue

Qualification scores flow back to the ad platforms as conversion signals — budget follows the campaigns that book calls.

Dedicated hiring

Add an engineer in a fortnight, not a quarter

A vetted developer working only on your product, on a monthly contract you can end. You run the interview and the standup. We carry the hiring, the payroll, the cover when someone is ill, and the replacement if the fit turns out wrong.

One person, a squad, or extra hands inside the team you already have.

How dedicated hiring works
One full-time senior developer, per month Market benchmark
US in-house $12,000 – $20,000+
Code Visionary Services Delivered $3,000 – $7,000

2026 public rate benchmarks at roughly 160 hours a month. Not a quote — we price per role after a scoping call.

Flagship product · Aria, built with Asobr Real Estate (UAE)

Aria: our flagship product, qualifying every inbound lead before an agent picks up the phone

Aria began as client work and became a product we now deploy in weeks. A brokerage running portal, WhatsApp and web enquiries lost its best buyers to response time. Aria is the multilingual assistant we built for it that answers instantly, asks budget, area, timeline and financing, then hands a scored, summarised lead to the right agent.

Shipped in eleven weeks — assistant, agent console, CRM sync, and a reporting layer for the sales director.

Read the full story
83%

of inbound leads qualified automatically, no agent time

38s

median first response, down from 5h

31h

agent hours saved per week

More work

Nova Wealth AI finance platform interface

Nova Wealth

AI-powered personal finance dashboard — 10,000+ users in month one, and a Series A raised on the back of it.

 K-BID auction platform bidding screen

K-BID

Cloud-native architecture rebuild for a nationwide auction platform — real-time bidding under 100ms, nationwide scale.

 Brain Meets Bytes podcast platform interface

Brain Meets Bytes

Podcast and research-community platform for a neuroresearcher — publishing and discussion, built to scale.
Pulse · the working log

We publish the work while it is still going wrong.

Case studies are written afterwards, once everything has been tidied up. This is the log we keep as we go: the bug we chased for two hours, the approach we threw away, the fix that turned out to be in a file production never loads.

56 entries published · newest

Read the full log
  1. debug

    LinkedIn changed its composer overnight

    Going to post a launch video to LinkedIn, the automation that had been reliable for weeks fell over on the first step: the selectors it hangs on were all dead. LinkedIn had swapped its post composer to a TipTap and ProseMirror editor sometime overnight — the media button was relabeled to Media, the dialog wrapper the flow keyed off had vanished, and the whole DOM had shifted underneath it. We rebuilt the video-post flow against the new structure and got the post out. Nothing about the task changed; the page it runs against did, without warning, which is the standing cost of automating somebody else's UI.

    Worked on by: Engineering
    Project: Growth Systems
  2. learning

    A banned number taught us where the line was

    One of our WhatsApp numbers got banned after bulk outreach went out through the CRM's send endpoint, which was backed by an unofficial adapter that talks to WhatsApp the way a phone would rather than through a sanctioned API. That is exactly the pattern the platform hunts for, and unmanaged volume through it is a number waiting to be cut off. We put a kill-switch on that endpoint so it now returns a 403 for WhatsApp while email bulk keeps working untouched, and decided to hold WhatsApp outreach until it can move onto the official Cloud API. The lesson was less about the ban than about not running outreach on a channel we didn't actually control.

    Worked on by: Engineering
    Project: Real-Estate CRM
  3. learning

    The habit we fixed was our own note-taking

    The recurring failure was never the code — it was losing the reasoning around it. A decision would get made in a chat, the session would reset, and the why behind it was gone by the next morning, so the same ground got re-argued from scratch. The fix was boring and it stuck: write every concrete decision to a dated file the moment it happens, then fold the durable ones into long-term notes on a slower loop.

    Treating our own memory as something to maintain, not something to trust, turned out to matter more than any single feature. It is cheaper to write the note than to rediscover the answer.

    Worked on by: Engineering
    Project: How We Ship
  4. debug

    An id with an at-sign that never survived the URL

    Opening a conversation that had come in over WhatsApp threw a 404 every time, while the ones from other channels opened fine. The lead id for those threads carries an @lid suffix, and it was going into the route still percent-encoded as %40, so the lookup was for an id that did not exist. Decoding it back before the route resolved made the thread open. It was the one character in the id that the URL layer wouldn't leave alone.

    Worked on by: Engineering
    Project: Real-Estate CRM

How we work

Four weeks to something real in production. No six-month discovery phase.

01

Discover

One week mapping your lead flow, the drop-offs, and the metric we'll be judged on.

02

Design

Conversation design, data model and interface prototyped and signed off before code.

03

Build

Two-week increments with evaluation sets, so quality is measured and not guessed.

04

Scale

Monitoring, retraining and new channels — or a clean handover to your team.

Client stories

Straight from the people who commissioned the work.

The people on your project

The five people who scope the work also build it.

Pradip Choudhury, COO of Code Visionary Services

Pradip Choudhury

COO

Shashank Singh, CTO of Code Visionary Services

Shashank Singh

CTO

Vikash Jha, CEO of Code Visionary Services

Vikash Jha

CEO

Sahil Biswas, CCO of Code Visionary Services

Sahil Biswas

CCO

Sayan Choudhury, CMO of Code Visionary Services

Sayan Choudhury

CMO

FAQs

Questions we get on the first call

If yours isn't here, ask it live — the call is twenty minutes and there's no deck.

Book a call →
How long until something is live?

Four weeks to a first production milestone on a qualification engagement. Larger platforms run in two-week increments after that, each one shippable.

Who owns the system you build?

You do — your cloud accounts, your repositories, your data. Aria is our product, but every deployment is configured and hosted for you, with an exit path documented from day one.

Do we need clean data before starting?

No. Discovery week includes an honest look at what you have. If the data won't support the outcome you want, we say so before you spend on a build.

How do you charge?

Fixed price for the first milestone so you can judge us on delivery, then monthly retainer for continued build, monitoring and paid-media work. No lock-in beyond the current month.

What if AI isn't the answer for us?

Then we tell you on the call. Plenty of lead problems are routing or follow-up problems, and those are cheaper to fix. We would rather lose the project than sell you a model you don't need.

Ready to stop losing leads?

Twenty minutes, one engineer, a straight answer on whether AI moves your numbers.

Book a call