Skip to content

Technologies

Engineering stack for products that have to work.

Backend systems, SaaS platforms, mobile products and cloud infrastructure. We choose the stack after the problem is clear — and every page here tells you when it is the wrong choice too.

Stack selector21 candidates

The brief

Client portal. Roles, documents, and a schema still moving.

Chosen

MERN Stack

One language end to end. The same API feeds a mobile app later.

Ruled out

A CMS build

Workflows are the product here, not content.

The brief

Public product pages. Organic search is the growth channel.

Chosen

Next.js

Server-rendered HTML. Core Web Vitals as a requirement, not a retrofit.

Ruled out

Client-side SPA

Google should not have to run your app to read it.

The brief

iOS and Android. One finite budget, parity expected.

Chosen

React Native

90%+ shared. One team ships both stores in one cycle.

Ruled out

Two native teams

Twice the cost for parity you would still be chasing.

The brief

Eligibility rules. Audited, provable, zero tolerance for a guess.

Chosen

Postgres + rules engine

Deterministic problem. Correct code, every time, for a millionth of the cost.

Ruled out

An LLM

The fashionable answer. We talk clients out of this one weekly.

  • React.js
  • Next.js
  • React Native
  • Node.js
  • Python
  • FastAPI
  • Cloud & DevOps
  • MERN Stack
  • MEAN Stack
  • MongoDB
  • PostgreSQL
  • Redis
  • Vector Databases
  • AI & LLM Engineering
  • LangChain
  • LlamaIndex
  • Haystack
  • CrewAI
  • LiteLLM
  • vLLM
  • Ollama

The stack is the answer. It is never the question.

  • React
  • Next.js
  • Node.js
  • TypeScript
  • React Native
  • Flutter
  • MongoDB
  • PostgreSQL
  • Redis
  • Express
  • NestJS
  • GraphQL
  • AWS
  • Azure
  • GCP
  • Docker
  • CI/CD
  • RAG pipelines
  • Vector search
  • LLM evals

Architecture first

Technology decisions connected to architecture.

A stack is only as good as the architecture underneath it. Every technology below is applied through the same delivery discipline.

  • 01

    Scalable APIs

    Secure APIs with performance and observability built in.

  • 02

    Event systems

    Queues, retries, async jobs and resilient flows.

  • 03

    AI infrastructure

    RAG, vector search, orchestration and guardrails.

  • 04

    Distributed data

    Models, caching, analytics and data pipelines.

  • 05

    Cloud delivery

    CI/CD, containers and infrastructure as code.

Technology fit guide

Which technology fits which business need?

Read it by what you are building, not by what is popular. The last column is the one most agencies leave off.

What you're buildingWhere we'd startWhyWhat we'd talk you out of
Client portal or admin dashboardMERN StackRoles, forms and data views in one connected JavaScript stackA CMS — workflows are the product, not content
Marketing site where search drives revenueNext.jsServer-rendered HTML and Core Web Vitals as a requirementA client-side SPA Google has to execute to read
App on both iOS and AndroidReact NativeOne codebase, 90%+ shared, one team shipping both storesFunding two native teams for feature parity
Ledger, payments or eligibility rulesPostgreSQLConstraints and transactions enforce correctness in the databaseA document store bent into relational shape
Catalogue or product data still changing shapeMongoDBDocuments model how the product actually reads its dataA migration every sprint to keep a schema in step
Answers over your own documentsVector search + LLMRetrieval grounded in your content, with citations and evalsA chatbot with no retrieval, guessing confidently
Slow pages under real trafficRedisCaching and queues absorb load before it reaches the databaseScaling the database to serve the same query repeatedly
An assistant that takes actions, not just answersLangChainTool calling and multi-step control flow you can read and testA framework wrapped around what is really one prompt
AI where data cannot leave your networkvLLMOpen-weight models served on your own infrastructureBuying GPUs before the volume maths says to
AI spend rising with nobody able to attribute itLiteLLMRouting, per-team budgets and usage visibility in one placeA proxy in front of a single model doing a single job
Python AI capability behind a product APIFastAPITyped contracts and generated docs at the service boundaryA second runtime added where Node would have done
Deterministic rules, audited and provablePlain code + PostgreSQLCorrect every time, for a fraction of the running costAn LLM — the fashionable answer, and the wrong one here

All technology pages

Explore each stack or tool in more detail.

21 pages, each connecting the technology to real service work and real engagements. Every one of them carries:

  • Practical selection guidance
  • Where it makes sense
  • Where we've shipped it
  • Pairings & services

Technology group

Applied AI and LLM engineering

AI shows up twice here: in the products we build with it, and in how we build everything else. Both are engineered, reviewed and measured rather than trusted. The frameworks below are the parts we assemble from — chosen per problem, and named here with the cases each one is wrong for.

AI / LLMFeatured

AI & LLM Engineering

RAG, agents and tool use built properly — with the eval harnesses and guardrails that keep them from embarrassing you.

LLMs · RAG · Agents · Evaluation harnesses

Explore AI & LLM Engineering
AI framework

LangChain

The orchestration layer for LLM applications — chains, tool-calling agents and memory, wired to your data and your systems.

Chains · Agents · Tools · Memory · Retrievers

Explore LangChain
AI framework

LlamaIndex

Retrieval built properly — ingestion, indexing and query engines tuned so an assistant answers from your documents instead of guessing.

Ingestion · Indexing · Query engines · Rerankers

Explore LlamaIndex
AI framework

Haystack

Composable search and RAG pipelines for enterprise estates — explicit, inspectable stages rather than one opaque call.

Pipelines · Retrievers · Rankers · Readers

Explore Haystack
AI framework

CrewAI

Multi-agent orchestration — specialised agents with defined roles handing work between them, for processes a single prompt cannot hold.

Agents · Roles · Tasks · Delegation

Explore CrewAI
Model routing

LiteLLM

One interface across every model provider — with routing, fallbacks, per-team budgets and the spend visibility that makes AI costs governable.

Unified API · Proxy · Budgets · Usage tracking

Explore LiteLLM
Model serving

vLLM

High-throughput inference for open-weight models — serving your own LLMs at production concurrency, on your own hardware.

PagedAttention · Batching · OpenAI-compatible API

Explore vLLM
Model serving

Ollama

Open-weight models running locally — for private workloads, air-gapped environments and development that costs nothing per token.

Local runtime · Model library · OpenAI-compatible API

Explore Ollama

The constant

What every stack here gets.

The technology is the variable. This is not — whichever of the nine above turns out to be right for you.

Every build gets

the same engineering discipline

  • Typed end to endTypeScript by default
  • Senior sign-offA named human on every diff
  • Tests to criteriaWritten to week-one acceptance
  • DocumentationSo the next team can read it
Shipped through

automated delivery, from week one

  • CI/CDDeploy on every merge
  • Infrastructure as codeEnvironments reproducible
  • StagingClickable from week three
  • ObservabilityTracing, alerting, dashboards
Hardened for

what actually breaks in production

  • Security reviewAdversarial review, secret scanning
  • Performance budgetsLatency and cost, costed up front
  • AccessibilityEmpty states and screen readers
  • Error trackingCaught before it is your incident
Handed over as

yours, with nothing to buy back

  • Your repositoryYour account from week one
  • Your cloudNo platform underneath
  • Your IPAssigned in the contract
  • No lock-inNothing to license back

Related services

Technology only matters if it supports the right service outcome.

Nobody buys a stack. These pages are where the technology choices above connect to what you are actually trying to get done.

FAQs

Questions about the technologies we use.

How stacks get chosen, what we work with, and who owns what at the end of it.

How do you choose a technology stack?

From the problem, in this order: what the product must do, who maintains it, whether search matters, how the data is really shaped, and what your team can hire for. The stack is the last decision, not the first — and we will recommend one we do not sell if it fits you better.

What technologies does TrivialWorks work with?

React, Next.js and React Native on the front; Node.js on the back, with Python and FastAPI where AI and data work make them the better tool; MERN and MEAN as connected stacks; MongoDB, PostgreSQL, Redis and vector stores for data; AWS, Azure and GCP underneath; and applied AI — LangChain, LlamaIndex, Haystack, CrewAI for orchestration and retrieval, LiteLLM for routing, vLLM and Ollama for self-hosted serving. Each has its own page here, including when not to use it.

Can you work with the stack we already have?

Yes, and often that is the honest recommendation. We inherit React, Node, PHP and legacy estates regularly. We audit first, stabilise, then extend — a rewrite gets proposed only when we can show you why the maths works, not by default.

Do you use AI to write our code?

Yes, through a six-stage pipeline with machine review and two mandatory human gates. No generated code reaches your main branch without a named senior engineer reading it and taking responsibility for it. The speed comes from automating volume, not from removing review.

Who owns the code and the IP?

You do, from day one — assigned contractually, in your repository and your cloud account throughout. There is no proprietary platform underneath and nothing to license back.

Can you recommend a stack before we commit to a project?

Yes. Send the requirement and you get back a functional specification — screens, data model, stack and an estimate — at no charge. If the honest answer is a smaller build than you asked for, that will be in there too.

Need help choosing the right stack?

Tell us the problem. You get back the honest shortlist — including the case for not building it the obvious way — from an engineer, not a salesperson.