onecandle

independent engineering studio

Better to light a candle.

Software, data, machine learning, and infrastructure. Shipped instrumented. Built to last.

Start a conversationWhat we do

“It is better to light a candle than to curse the darkness.”

// services

A small studio, a wide bench.

Engagements are scoped to what you need: a one-week spike, a multi-month build, or an ongoing retainer. Below is where we work.

  • Full-stack engineering

    Greenfield products and pragmatic upgrades. Web, APIs, and data layers, running wherever you need them: cloud, on-prem, hybrid, public, or sovereign. The stacks below are where we've shipped, not where we're capped. Comfortable with the codebase you have, and the hardware underneath it.

    • React, Next.js, Svelte, vanilla
    • Node, Python, Go, Elixir, Rust
    • Postgres, Redis, & friends
    • Polyglot by default

  • Data engineering

    Pipelines, warehouses, and lakes that hold up. Batch and streaming, ETL and ELT, with the schemas and monitoring that turn raw data into something you can actually query.

    • Postgres, ClickHouse, dbt
    • Kafka, Spark, Airflow
    • Schema design & quality

  • Machine learning & deep learning

    Modeling for real use: classification, regression, forecasting, and deep learning. Shipped with the evaluation and observability that keeps models trustworthy in production.

    • PyTorch / TensorFlow
    • Fine-tuning & embeddings
    • NLP, CV, time-series

  • LLM & agent systems

    LLM-powered products and the plumbing behind them: agent workflows, tool integrations, retrieval and long-term memory, and the evaluation harnesses that tell you whether any of it actually works.

    • Agents & MCP servers
    • RAG & long-term memory
    • Evaluation & guardrails

  • Analytics & decision support

    Statistical modeling, experiment design, forecasting, and BI. Operational data translated into decisions you can defend, with the methodology to back them.

    • A/B testing & causal inference
    • Forecasting & operations research
    • Dashboards & metrics

  • Research & prototyping

    Honest exploration of an idea. We build the smallest thing that proves or disproves it, then document what we learned.

    • Feasibility spikes
    • Technical due diligence
    • Proof-of-concepts

  • DevOps & infrastructure

    Reproducible environments, predictable deploys, and observability that's actually useful. Cloud, self-hosted, or fully sovereign, with the uptime discipline production demands.

    • CI/CD & IaC
    • Containers & Kubernetes
    • Uptime & reliability

  • Security & risk readiness

    Pre-mortems and hardening for teams that need to get it right the first time. We map a system's failure modes: what's exposed, what's brittle, what the blast radius is. We also set up key management and operating posture where uptime is non-negotiable.

    • Pre-mortems & failure-mode analysis
    • Code & infra review
    • Key management & HSMs

  • Advisory

    A senior set of eyes when you need one. Architecture review, technical due diligence, and honest assessment of where a project stands.

    • Architecture review
    • Roadmap pressure-test
    • Hands-on as needed

// about

Quiet, careful, built to last.

One Candle is run by Ian Culp. A senior engineer who runs toward complexity when the right solution lives there, bakes resilience into the work from the start, and plays devil's advocate so the trade-offs are visible before you decide.

The studio takes on a small number of engagements at a time so we can do them properly. Greenfield product work, R&D you can hand to your investors, infrastructure you can hand to your next hire.

Background: graduate training in analytics (M.S. Analytics, Georgia Tech), years of shipping production software and data/ML systems across web, infrastructure, and applied ML, plus several years operating high-uptime distributed systems in adversarial environments.

Resilience gets baked in from day one: what's exposed, what the blast radius is, what's logged. Dependencies you can defend, infrastructure you can audit, observability that catches what matters. Building with that posture early is much easier than retrofitting it under pressure later.

You stay current. Status updates without being asked, problems surfaced before they're emergencies.

based
Remote · U.S.
availability
Booking discovery calls

// contact

Tell us about it.

The best engagements start with a clear, honest description of what you're trying to do and what's in the way. A paragraph is plenty.

Prefer email? Reach us at hello@onecandle.dev.