The Stack That Powers Everything
Every layer of our technology — from pixel to database to cloud — chosen for a reason.
Frontend — What Users See
React 18
Our primary UI library — component model, concurrent features, and the broadest ecosystem of any frontend framework.
Next.js 14
App Router, RSC, and edge runtime. Full-stack React with built-in performance and SEO primitives.
Tailwind CSS
Utility-first CSS with a design token system. Fast to write, easy to maintain, no unused styles in production.
Framer Motion
Production-grade animation library. Smooth micro-interactions and page transitions that feel native.
Figma
Design system source of truth. Engineers consume tokens and components directly from Figma variables.
Vitest + Playwright
Unit and component tests in Vitest, full end-to-end browser automation with Playwright.
Vite
Lightning-fast HMR and optimised production bundling. Replaces Webpack for all new projects.
Storybook
Isolated component development and visual regression testing. Living UI documentation.
Backend — Where Logic Lives
Python 3.12
Primary language for AI/ML pipelines, data engineering, and automation services. FastAPI for async REST APIs.
PHP 8.3
Powers our CMS-driven and portal products. Clean OOP with strict typing and modern PSR standards.
Node.js / Express
Real-time features, WebSocket servers, and API gateways where low latency and high concurrency matter.
FastAPI
High-performance Python REST framework with automatic OpenAPI docs and Pydantic validation.
GraphQL / Apollo
Flexible data querying for complex product interfaces. Reduces over-fetching across mobile and web clients.
OAuth 2.0 / JWT
Token-based auth with refresh rotation, PKCE, and role-based access control across all services.
Celery + Redis
Distributed task queue for async jobs — email delivery, report generation, and scheduled data syncs.
REST + OpenAPI
All APIs documented with OpenAPI 3.1 specs. Client SDKs auto-generated for partner integrations.
Databases — How We Store It
PostgreSQL 16
Primary relational store. JSONB for semi-structured data, full-text search, and row-level security.
MongoDB
Document store for flexible schemas — content management, logs, and rapidly evolving data models.
Redis 7
In-memory caching, session management, pub/sub messaging, and rate limiting.
Elasticsearch
Full-text and vector search across large document collections. Powers all search features.
ClickHouse
Column-oriented OLAP for analytics dashboards. Sub-second queries on billions of rows.
Pinecone / pgvector
Vector databases for semantic search and RAG pipelines feeding our generative AI features.
Delta Lake
ACID-compliant data lake for ML feature stores and large-scale analytics pipelines.
DBeaver / Metabase
Internal data exploration and BI dashboards. Self-service analytics for non-technical stakeholders.
Cloud & DevOps — How We Ship It
AWS
Primary cloud provider. EC2, ECS, RDS, S3, CloudFront, and Lambda underpin our production workloads.
Google Cloud
Vertex AI, BigQuery, and Cloud Run for AI-heavy workloads and multi-cloud redundancy.
Docker
All services containerised. Reproducible builds from laptop to production with identical environments.
Kubernetes (EKS)
Container orchestration with autoscaling, rolling deployments, and zero-downtime updates.
GitHub Actions
CI/CD pipelines for every repo. Build → Test → Deploy triggered on every pull request merge.
Terraform
Infrastructure as Code. Every cloud resource version-controlled, reproducible, and auditable.
Grafana + Prometheus
Full observability stack — metrics, logs (Loki), traces (Tempo), and on-call alerting via PagerDuty.
Vault by HashiCorp
Centralised secrets management. Dynamic credentials and zero-trust access for all services.
AI & ML — The Intelligence Layer
PyTorch
Primary deep learning framework for custom model training, fine-tuning, and research experiments.
scikit-learn
Classical ML — regression, classification, clustering, and model selection utilities.
LangChain
Orchestration framework for LLM chains, agents, tool use, and RAG pipeline construction.
MLflow
Experiment tracking, model registry, and deployment lifecycle management across all ML projects.
Hugging Face
Model hub and Transformers library for NLP, vision, and multimodal model fine-tuning.
Pandas / Polars
Data manipulation and transformation. Polars replaces Pandas for large-scale performance.
Apache Airflow
Workflow orchestration for data pipelines, model retraining schedules, and ETL jobs.
OpenCV + YOLO
Computer vision pipelines for image preprocessing, object detection, and OCR automation.
Why These Choices Specifically
Every technology in our stack was chosen against a set of criteria, not just popularity.
Hiring Pool
Mainstream, well-documented technologies mean a wider talent pool and faster onboarding. Exotic choices create single points of human failure.
Proven at Scale
Every tool has been stress-tested at millions of requests per day by other organisations. We don't experiment with unproven tech in production.
Total Cost of Ownership
Licensing, hosting, operations, and migration costs are all considered upfront. We prefer open source where it doesn't add unacceptable operational burden.
Integration Surface
Every tool needs to integrate cleanly with the rest of the stack. Rich SDK ecosystems and open standards reduce glue code and maintenance drag.
Security Posture
Maintained projects with a history of rapid CVE patching, clear security disclosure processes, and no known long-standing critical vulnerabilities.
Learning Resources
Good documentation, active communities, and accessible learning resources. A tool our team can master — not just use — is a strategic asset.
Building with a Similar Stack?
Whether you need to modernise an existing system or start fresh, our engineers know this stack inside out.