Who you are hiring
An engineering studio building AI-driven products, automation systems and trading infrastructure for founders and ops leaders worldwide.
Founded in December 2022 by Selvaraj Kalimuthu, who spent 15+ years on production systems before starting it: the matching engine, trade executor and order-management system at Multipli.fi, technical lead at the crypto exchange WazirX and at ShopX, and consulting work before that at Urban Ladder and CaratLane.
The studio
THWorks has been an engineering studio since December 2022, working remotely with clients worldwide.
Running our own products is not a side project. It is where we find out what our own advice costs: a bad architectural call on a client build is an invoice and an apology, and on ours it is a support queue we have to sit in ourselves. Most of what we refuse to do to a client system, we learned by doing it to one of these.
Backtest a strategy, paper-trade it, then run it live on the same engine, across eight brokers.
Read about TraderOpsSales automationScrape leads from Google Maps or YouTube, audit their website, and put an AI agent on the phone to pitch them.
Read about OutreachKitsStartup toolingScores a startup idea 0–100 across five dimensions, from market viability to defensibility. Free to use.
Read about IdeaValidatorWhy we build this way
We price the job, not the hours, and the number is agreed before anyone starts. So when a build goes faster, the saving lands on your side of the invoice instead of ours.
Scaffolding, wiring, the fourth CRUD screen, the integration you have written nine times before — that was most of a build once, and most of the invoice. We generate that layer now. The hours it frees go to the parts that decide whether the thing actually works: the data model, the failure paths, and anything with money behind it.
The other half is what people are right to worry about. A model will produce code that runs and is wrong, cheerfully, and it will not tell you which kind it just wrote. So nothing lands in your repository because a model produced it. It lands there because an engineer here read it, understood how it behaves under load, and would defend it in review. That standard came out of matching engines and a crypto exchange, where a bad merge costs somebody real money rather than an apology.
Hourly billing quietly rewards the slow version of a project. We would rather not be paid for that.
Illustrative
How a change reaches your repository
Scaffolding, wiring, boilerplate, and the integration shape we have written before.
An engineer here reads it. Not skimmed for style — read for what it does under load.
The data model, the failure paths, and anything with money behind it get argued about.
It lands only once somebody here would defend it in review.
By the numbers
15+
Projects delivered
4+
Years in operation
10+
Countries we serve
3+
In-house products
What we hold to
Trading-operations software does not have to look like 2008. We build interfaces traders and ops teams actually want to use.
AI agents, n8n / Make / Zapier workflows, and RAG systems baked into the products we ship — not bolted on.
Zero-tolerance discipline carried over from matching engines and crypto exchanges: the failure paths get built before the happy path counts as done.
Clean code, sane defaults, debuggable systems. Built so the next engineer to touch it can ship without re-learning.
Selected work
Five engagements, with the figures we measured on the way out. Every number here also appears on its own case-study page.
01
Enterprise support
Peak-hour call abandonment sat at 35%. A voice agent now answers in under two seconds, resolves most calls outright, and hands the rest over with context.
Read the case study
82%
Calls resolved without a human
1.8s
Response latency
4.5K+
Appointments booked monthly
02
Fintech
Bot attacks had climbed 400% while engagement fell 20%. Multimodal support now defends the funnel and answers in under half a second, with no extra step for real users.
Read the case study
99.2%
Bot defence accuracy
60%
Faster voice response
87%
Onboarding completion
03
Real estate
Leads were waiting four hours for a callback. Every enquiry is now engaged within ten seconds, booked into the calendar, and handled without a coordinator.
Read the case study
42%
More property tours booked
10s
Lead response time
5,500$+
Saved monthly
04
Global retail
Churn in non-English markets ran at 22%, driven by cultural misreads. Sentiment is now mapped per locale, so the cases that need a human reach one.
Read the case study
95%
Escalation detection precision
89%
Sentiment accuracy
18%
Churn reduction
05
B2B SaaS
Ten hours a week were going into manual Reddit prospecting. A multi-tenant scraper now watches 1,000+ subreddits hourly and scores what it finds.
Read the case study
75%
Infrastructure savings
5K+
Leads captured daily
2min
Time to onboard a user
How we work
Four steps, no discovery theatre. You see working software in the first fortnight and you own every line of it.
Scope in one week. Fixed price, fixed length, and the spec is yours to keep whether or not you build it with us.
architecture
spec.md
yours to keepyour repo
liveBuild in the open. Your repo and your cloud account from day one, with a working demo every Friday.
pre-launch
p99
00ms
Harden before launch. Tests, observability, failure paths and rate limits — the half where trading systems and agents actually break.
Run it with you. We stay on through the first months, then hand over with documentation written for engineers, not for a signature.
About us
Where we are, how we work with teams in other timezones, and how to reach someone.
An engineering studio. The distinction we hold to is that a senior engineer is close to every problem — the person who scopes your build is one of the people who writes it, and there is no layer between you and them whose job is to relay. It is why we stay deliberately small.
Bengaluru, India — and we work remotely with founders and teams worldwide.
Asynchronously by default, with a fixed overlap window agreed at the start of the engagement rather than negotiated weekly. Progress is visible in the repository and in a written update on a schedule you set, so the question "what happened yesterday" never needs a meeting to answer.
Because the work is mostly systems that have to keep running — execution, reconciliation, agent loops with real money or real customers behind them. That is a different discipline from shipping a campaign site, and staffing it like one is how those projects fail. It also means we turn down work that would be better served by an agency.
Email contact@thworks.org, or selva@thworks.org to reach the founder directly. Call or WhatsApp +91 97883 43294. We reply within one business day, including to say no.
Next step
A scoping call is 30 minutes and gets you a fixed estimate. If we are the wrong people for it, we will say so.
