Where the latency actually goes
A tick-to-order budget for a NIFTY weekly options engine on a retail broker API, and why the broker leg, not your code, sets the floor.
5 min read
Low-latency trading systems
NSE · BSE · MCX · Crypto
Everyone sells algo trading. We build the low-latency system underneath it.
An algo is a rule. A system is the feed, the book, the order path and the exits that carry that rule from tick to order in under a millisecond, and through a whole live session without dropping one. That low-latency layer is the part nobody else is selling, and it is the only part we build.
100+
Brokers supported
200-level
Depth, decoded
<1 ms
Tick to intent
Why us
Production execution without the production surprises.
Every algo shop can show a backtest. What decides whether a system survives a live session is smaller and less glamorous, and it is what we design for first.
-
Under 1 ms
Tick to intent, in process.
Socket read, book update, analytics, decision and risk gate all run in memory. Nothing on that path touches a disk, a lock or the network. The broker leg is the slow part, and the engine is built never to be the slow link behind it.
-
One cascade
Every exit, every strategy.
Stop, target, trail, time limit and expiry buffer evaluate in one fixed order for every open position. A strategy writes a plan and hands off. The engine sells.
-
Byte for byte
Yesterday's session, replayed exactly.
The recorder keeps the raw wire bytes, not a parsed summary. A session replays with order placement off, and its decision log is hashed and diffed against the live run.
-
Stops itself
Six independent ways it refuses to trade.
A daily loss floor, a consecutive-loss switch, a broker-reject cooldown, a stale-feed ladder, a data-feed fatal and a funds check. Each one is independent of the others, each writes its reason to the ledger, and each halts the engine before an operator has to.
End to end
Design, screen, engine, data, and the box it runs on
Five stages, one team. Keep scrolling.
Dashboard
Engine
Data
Box
Broker APIs
Any broker · any venue
Built against the brokers your desk already uses
Point at any planet to see what that broker's API gives you: depth levels, order limits, brokerage and the API fee, from their own documentation.
Zerodha KiteUpstoxDhanAngel OneFyers5paisaICICI DirectKotak NeoHDFC SkyMotilal OswalIIFL SecuritiesAlice BlueShoonyaPaytm MoneyGrowwSharekhanNuvamaChoiceSamcoFlattradeJainamBigulFirstockZebuMaster TrustGoodwillTradejiniProStocksTradeSmartAxis DirectSBI SecuritiesAnand RathiGeojitReligareVenturaStoxkartm.StockBajaj BrokingRupeezyIndira SecuritiesSAS OnlineLemonnINDmoneyJM FinancialNirmal BangEspresso
NSEBSEMCXNCDEXGIFT NiftySymphony XTSTrueDataGlobal DatafeedsDhan Data APIKite ConnectDelta ExchangeCoinDCXBinanceBybitOKX
Any
broker API your desk already uses
3,000+
strategy candidates tested before one ships
200
depth levels decoded, every tick
<1ms
tick to intent, in process
Who we build for
Three kinds of desk, one kind of problem
You have the capital or the edge. The system underneath is the part we build and hand over.
Prop desks
Your capital, your rules, our execution core.
Low-latency pathMulti-strategyRisk gate
Institutions and funds
An order path your compliance function can read.
AuditableReplayableDocumented
Traders with an edge
A method that works by hand, run without you.
AutomationExitsPaper first
What we build
Eleven systems, one spine
Every engagement is some combination of these. Hover one.
Low-latency execution engine
Book decoding to submission on one hot loop.
C++17CythonLock-free
Strategy automation
Your rules, running unattended, the same on day forty.
on_tickRegime routerKill switch
Real-time market scanner
Every strike on every tick, filtered to the few that matter now.
OI shiftsIV rankSweepsAlerts
Copy trading and replication
One decision, many accounts, each with its own caps.
Master and followersPer-account sizingTagged
Multi-broker integration
Each venue on its own adapter, capabilities declared at boot.
DhanUpstoxFyersAngel One
Option chain and OI analytics
A live chain with your own Greeks and OI boards.
Black-76OI by strikePCR
Backtesting and research
Yesterday's wire bytes, replayed and diffed.
Wire captureReplaySealed holdout
Risk and compliance layer
One gate before the OMS; floors that scale with capital.
R:R floorDaily floorAudit trail
Operator dashboards
Health lamps, the position book, the ledger, the log.
Next.jsWebSocketGrafana
Crypto execution
The same engine, a 24/7 session and a funding-aware cost model.
Spot and perps24/7 clockOne adapter
Rescue and audit
A system somebody else built, read, fixed and documented.
ForensicsReplay the bad dayRunbook
Architecture
Tick to trade, the whole path
A tick leaves the exchange and comes back as a fill. Scroll it through. Point at the book. Send an order.
Station 1 of 8
Exchange
A tick leaves the matching engine.
Order management
An order is a state machine, not a REST call
place_order() returns an acknowledgement, sometimes. Scroll the order through.
NEWSUBMITTEDACKNOWLEDGEDIOC GRACEFILLED
NEW
Accepted by risk. An ID is minted before anything leaves the process.
Idempotent client IDsTranche-aware fillsUNCERTAIN is a stateDurable per transition
·
Why this practice
What goes wrong with the other quote
We hear these five complaints most often about the last vendor, and each one is something we handle differently, on purpose.
Selected work
Systems that have taken a live session
Intraday options engine
A production intraday options engine on a 200-level depth feed. It streams NIFTY and BANKNIFTY weeklies, routes between strategies on a regime classifier and manages every exit through one cascade. Deterministic replay, a Postgres ledger and Prometheus metrics, so any session can be reproduced tick for tick months later.
Private trading desk & control plane
A complete desk for one client. Engine, web control plane and eleven operator screens. He pastes his broker token, presses start, and watches the chain, the OI boards and the decision tape. Paper and live are separate rungs, and the switch between them is a deliberate act with a confirmation.
Writing
Latest from the blog
Notes from building execution systems for Indian options desks. Written for the engineer who has to run one, not for the search engine.
Replaying a live session byte for byte
Record the wire bytes, not the parsed ticks, hash the decision log, and never grade a change on the first half of a tape.
5 min readExits, not entries, decide whether an algo survives
One exit cascade for every strategy, a fixed evaluation order, and a risk gate that sits in front of the OMS instead of inside any strategy.
5 min read
Start a project
Tell us what breaks
The more specific you are, the more useful the first reply. Broker, product, and what the current system does wrong beats "I want an algo".
Good fits
Prop desks needing an execution or risk layer · traders with a proven manual edge to automate · fintech teams who need an order path they can trust · anyone holding an engine they cannot debug.
Not a fit
"Build me a strategy that makes X% a month." We build systems that execute your strategy correctly. We do not sell signals, tips or returns.
- Based in
- India, working in IST
- Building for
- Prop desks, funds and independent traders
- Stack
- C++, Python, Postgres, Next.js
