OUR WORK

Work we can talk about

Five engagements, each with the constraint we were working against, the decisions we made, and how the system holds up under load. There is more under NDA.

Fliks — a streaming platform that had to feel instant
  • IndustryStreaming media
  • ServicesWeb · Cloud
  • Duration6 months
  • StackReact · Node.js · HLS · AWS
FLIKS · FOR YOURBACKERS
A streaming platform that had to feel instant

THE CHALLENGE

YourBackers needed a video platform with live delivery, content management and multi-device playback — launched on a fixed date, with buffering treated as a bug, not a network fact of life.

WHAT WE DID

  • Adaptive delivery first. HLS with per-segment bitrate switching, so playback degrades to a lower resolution instead of stalling.
  • A CMS the editors could run. Publishing, scheduling and access control without an engineer in the loop.
  • One player, every screen. A single web player abstraction across desktop, mobile web and TV browsers.

SECURITY & SCALE

Signed URLs and short-lived tokens on every stream; origin shielded behind a CDN. Load-tested to [TO CONFIRM] concurrent viewers before launch.

[TO CONFIRM] ACTIVE USERS · FIGURE HELD PENDING THE APPROVED CLAIMS LIST
Read the full case study
SHOPLYTICS — PROJECT THUMBNAIL Drop a wide product image or screenshot here (real UI, navy field, soft shadow).
  • IndustryDigital retail
  • ServicesWeb · SaaS
  • Duration5 months
  • StackNext.js · PostgreSQL · Redis
SHOPLYTICS
Headless commerce with inventory that never lies

THE CHALLENGE

A retailer selling across several warehouses was overselling stock during peaks. The commercial constraint: fix it without a replatform and without checkout downtime.

WHAT WE DID

  • Inventory as the source of truth. Real-time multi-warehouse sync with reservations, so two carts can't claim the last unit.
  • Headless, storefront-agnostic. A commerce API the existing front end could adopt incrementally.
  • One-click checkout. Tokenised payment and saved addresses to cut the drop-off at the final step.

SECURITY & SCALE

PCI-compliant payment handling, no card data on our servers. Order processing held its success rate across [TO CONFIRM] peak-season transactions.

[TO CONFIRM] GMV GROWTH · RELATIVE MEASURE PENDING CLIENT CONSENT
Read the full case study
BLOWISTLE — PROJECT THUMBNAIL Drop a wide product image or screenshot here (real UI, navy field, soft shadow).
  • IndustryAthletic tech
  • ServicesMobile · AI/ML · Cloud
  • Duration8 months
  • StackPython · Kafka · MongoDB
BLOWISTLE · OUR OWN PRODUCT
Sports analytics from a booking to a scorecard

THE CHALLENGE

We wanted athletes to train against data, not memory: book a session at a partner venue, get it recorded, and receive analytics afterwards — end to end, owned by us.

WHAT WE DID

  • Booking before cameras. The scheduling and venue layer shipped first, so the pipeline always had real sessions to process.
  • A streaming analytics core. Kafka ingest with per-event processing rather than overnight batch, so results land while it still matters.
  • Predictions with a baseline. Every model states the task and the baseline it beats — no accuracy figure without both.

SECURITY & SCALE

Athlete footage is private by default, access scoped per session. Built to add venues without re-architecting ingest.

[TO CONFIRM] DATA POINTS/DAY · MUST RECONCILE WITH SESSIONS RECORDED
Read the full case study
ARCHIGEN — PROJECT THUMBNAIL Drop a wide product image or screenshot here (real UI, navy field, soft shadow).
  • IndustryGenerative AI
  • ServicesAI/ML · Web
  • Duration4 months
  • StackPython · PyTorch · Three.js
ARCHIGEN
Floor plans and 3D from a sentence

THE CHALLENGE

Turn a plain-language brief — rooms, adjacencies, constraints — into a valid floor plan and a 3D visualisation, fast enough to iterate on in a meeting.

WHAT WE DID

  • Constraints as hard rules. A solver enforces adjacency and code limits so generated plans are buildable, not just plausible.
  • Generation, then refinement. The model proposes; the user nudges; the plan updates without starting over.
  • Web-native 3D. Real-time visualisation in the browser, no plug-in, no export step.

SECURITY & SCALE

Per-project isolation; generated assets are the client's alone. Inference queued and autoscaled to keep interactive latency stable.

[TO CONFIRM] DESIGNS GENERATED · FIGURE HELD PENDING THE CLAIMS LIST
Read the full case study
QUANTEDGE — PROJECT THUMBNAIL Drop a wide product image or screenshot here (real UI, navy field, soft shadow).
  • IndustryQuant trading
  • ServicesCloud · Systems
  • Duration10 months
  • StackRust · Go · kdb+
QUANTEDGE
A trading platform where latency is the product

THE CHALLENGE

Custom algorithms, real-time market analysis and risk controls — on a system where every microsecond on the tick-to-trade path is a competitive cost.

WHAT WE DID

  • Rust on the hot path. The latency-critical route in Rust; orchestration and tooling in Go where developer speed matters more.
  • Risk in-line, not after. Pre-trade risk checks inside the order path, so a bad order is stopped before it leaves.
  • Measured, not asserted. The tick-to-trade path is instrumented end to end, and we state exactly what the number includes.

SECURITY & SCALE

Least-privilege access to market connectivity, full audit trail on every order. Capacity-tested against historical peak message rates.

[TO CONFIRM] TICK-TO-TRADE · MEASURED PATH TO BE SPECIFIED
Read the full case study

FLAG, DON'T FABRICATE — EVERY FIGURE ABOVE IS HELD AT [TO CONFIRM] UNTIL IT APPEARS ON
THE APPROVED CLAIMS LIST. NO CLIENT DATA IS PUBLISHED WITHOUT WRITTEN CONSENT.

GET IN TOUCH

Want the version with the numbers in it?

Book a call and we'll walk you through a relevant build — the constraint, the decision, and the figures we can share under NDA.

Book a 30-minute call

REPLY WITHIN ONE BUSINESS DAY · KOVILPATTI, IST