Gold Maple Tree
—  Capabilities

Four disciplines, one practice.

Each discipline holds its own. We are most useful where they meet — when an engineering question has a marketing answer, when a marketing question has a blockchain answer, when the right model unlocks a workflow that no human could carry alone.

How to read this page

Each pillar declares what we deliver, how we engage, the stack we lean on, and a point of view. We have also written down what we do not do, because telling you up front is cheaper for everyone.

Pillar I

Technology

Engineering as inheritance, not dependency.

We engineer for companies that intend to operate the system after we leave. Most of what we ship is small, durable, and self-explanatory; we are wary of architectures that require us to be present. The codebase you receive should read more like a careful book than a haunted house.

— What we deliver
  • Production web platforms (Astro, Next.js, server-rendered)
  • API & integration backbones — REST, GraphQL, signed webhooks
  • Cloud infrastructure provisioned in code (Terraform, Pulumi)
  • Postgres schema design, migrations, observability, backups
  • Internal tools that quietly retire spreadsheets
  • Hardening, audit prep, dependency hygiene, supply-chain review
— How we engage
  • Audit
    2 weeks · fixed

    We read the codebase, the infra, and the runbook, then write the truth.

  • Sprint
    4–8 weeks

    A discrete piece of work, scoped to a single shipping moment.

  • Embedded
    3–6 months

    We sit inside your team and build with your engineers, not around them.

— Tools we lean on
TypeScript Python Go PostgreSQL DigitalOcean Terraform Cloudflare GitHub Actions Sentry
Not for

Throwaway prototypes that no one expects to maintain. There are faster firms for that.

The right time to refactor is rarely now — but the right time to design for refactoring is always at the start.

—  A point of view, Pillar I
Pillar II

Social Media Marketing

Modern social is infrastructure, not creative.

The accounts that win in 2026 are not the ones with the best taste, but the ones with the best pipeline. A single editorial voice, fed by AI assistance, scheduled across YouTube, Instagram, TikTok, and X, with analytics returned cleanly enough to act on. We build that pipeline — and operate it for clients who would rather not.

— What we deliver
  • YouTube channel strategy: content frameworks, thumbnail and title testing, retention analysis
  • Short-form architecture for Instagram Reels and TikTok — hooks, pacing, cliffhanger maths
  • AI-assisted production: Whisper transcript → LLM rewrite → automated cut
  • Programmatic cross-platform publishing with shared metadata and tracking
  • Custom analytics dashboards that surface cause, not just effect
  • Operator playbooks so a single editor can run ten channels without burning out
— How we engage
  • Channel audit
    1 week · fixed

    We dissect the last 90 days of output and tell you what's actually working.

  • Pipeline build
    4–6 weeks

    We wire up the publishing, scheduling, and analytics infrastructure end-to-end.

  • Operator
    Ongoing retainer

    We run the social arm of the business. You read the weekly memo.

— Tools we lean on
YouTube API Meta Graph API TikTok API Whisper Claude GPT-4 Descript ffmpeg n8n Notion
Not for

Single-platform vanity. If you only want followers and don't care about what they do, look elsewhere.

Brand and performance are not opposing forces. They are the same lever, pulled at different time horizons.

—  A point of view, Pillar II
Pillar III

Blockchain

Considered, audited, deployable.

We work for institutions and founders who can afford no shortcuts — the kind of operator who would lose more in a single exploit than they would gain from cutting corners. Our blockchain practice is small on purpose: senior, slow, and adversarial about its own work. If you need a token in two weeks, we are not the right firm.

— What we deliver
  • Smart contract design, implementation, and pre-deploy review
  • Tokenisation strategy: economic design, distribution, vesting
  • Wallet integrations across EVM chains, Solana, and Bitcoin
  • On-chain analytics: pipelines from raw chains to operator dashboards
  • Adversarial security review of existing contracts and protocols
  • Treasury, multisig, and operational infrastructure for on-chain organisations
— How we engage
  • Contract audit
    2–4 weeks

    Adversarial review of existing Solidity, Vyper, or Move with a written report.

  • Greenfield protocol
    8–16 weeks

    Economic design through to mainnet deployment, with the audit included.

  • Advisory
    Retainer

    Ongoing strategic and technical input for teams shipping continuously.

— Tools we lean on
Solidity Foundry Hardhat Viem OpenZeppelin Glassnode The Graph Tenderly Safe
Not for

Memecoins, airdrop farms, and protocols whose pitch deck has more graphs than its codebase has tests.

The block is the oldest cryptographic primitive. There is no shame in being slow about it.

—  A point of view, Pillar III
Pillar IV

Artificial Intelligence

Models are a feature. Workflow is the product.

We are uninterested in AI as a category. We are interested in the agent that retires a recurring task, the pipeline that generates a missing dataset, the assistant that quietly raises every employee's ceiling. Most AI projects fail not because the model is too weak — they fail because the workflow it was dropped into was never properly understood.

— What we deliver
  • LLM integration into existing products with versioned prompts and evaluations
  • Agent design: tool use, planning, retries, human-in-the-loop checkpoints
  • Data pipelines: extraction, cleaning, embedding, retrieval, evaluation
  • Targeted fine-tuning when, and only when, it materially helps
  • Prompt and evaluation infrastructure that survives model upgrades
  • Internal AI tooling for teams that already exist — no platform play required
— How we engage
  • Discovery
    2 weeks · fixed

    We identify which workflows AI changes meaningfully, and which it merely decorates.

  • Build
    6–12 weeks

    We ship the agent, the pipeline, or the integration — measurable, instrumented.

  • Operate
    Retainer

    We run evaluations, maintain prompts, and migrate across model generations.

— Tools we lean on
Anthropic Claude OpenAI Anthropic SDK LangGraph pgvector LangSmith Hugging Face Modal DigitalOcean
Not for

AI for AI's sake. If the operator can't articulate what success looks like in plain English, no model will save them.

Most AI projects fail not because the model isn't good enough, but because the workflow it was dropped into was never properly understood.

—  A point of view, Pillar IV
Process

How an engagement begins.

01 — Letter

You write to us — not a contact form, an actual email. A few paragraphs about your company and what you are working towards is enough. We respond personally within two business days.

02 — Conversation

One call, one hour. We listen more than we talk. By the end of it we will have agreed whether there is a fit at all, and if so, what shape the engagement should take.

03 — Memo

Within a week, a written proposal: scope, deliverables, timeline, fee, what is and is not included. Short, specific, signable. No surprises later.

04 — Work

We begin. Weekly written updates, a shared workspace, and access to the people doing the work — not an account manager standing between you and them.

Continue

Work.

An index of selected engagements.

Read on