Selected work

Products we built. Names we can't print.

Every engagement below is anonymised at the client's request. The problems, the builds and the numbers are exactly as delivered.

  • All
  • Social
  • Retail
  • Media and events
People laughing and talking together outdoors
Social · Dedicated team

Community interaction platform

A global consumer social network

Challenge

Conversation threads, groups and direct messaging had grown through separate product teams, leaving users with fragmented identity, inconsistent notifications and slow moderation response.

What we built

A unified social layer for feeds, messaging and shared-interest communities, with real-time event pipelines, trust-and-safety tooling and ranking controls for product and moderation teams.

+31%

Weekly interactions per user

-44%

Moderation response time

  • Social graph
  • Realtime events
  • Messaging
  • Trust and safety
Interior of a modern fashion retail store
Retail · Project delivery

Headless commerce replatform

A global fashion retailer

Challenge

A monolithic storefront buckled every peak season, capping revenue on the retailer's biggest sales days.

What we built

A composable, API-first storefront with edge rendering and an autoscaling checkout, rolled out market by market with zero downtime.

4.1x

Peak throughput

-52%

Page load time

  • TypeScript
  • Next.js
  • GraphQL
  • Edge
Crowd at a live concert with stage lights
Media and events · Dedicated team

Real-time demand engine

A global live-events and ticketing group

Challenge

Static pricing left money on the table and stranded inventory across thousands of venues with wildly different demand.

What we built

A streaming demand engine that reprices and reallocates inventory in real time, with guardrails the commercial team controls.

+22%

Revenue per event

<200ms

Repricing latency

  • Rust
  • Flink
  • Redis
  • AWS
Television screen in a living room setup
Media · Data, AI and ML

AI content personalization

A streaming media network

Challenge

A generic homepage meant viewers churned before finding anything worth watching.

What we built

A recommendation platform serving personalized rows to 40M monthly viewers, with an experimentation stack to test every ranking change.

+16%

Watch time

40M

Monthly viewers served

  • Python
  • PyTorch
  • Feature store
  • MLOps

Let's build

Your project could be the next one we can't name.

sales@think-digital.ge