No magic demos
We do not sell a slide of a chatbot. The first artefact you see is a process map with your own volumes and cycle times on it.
Studio · since 2019
A small studio that ships, and hands over.
We are deliberately small: four disciplines, one engineer embedded per project, and a delivery lead who is accountable to you every week. No account managers, no offshore handoff, no black box.
HELIXWORKS started in 2019 with a simple observation: most companies do not need another platform, they need the one process that eats their week to stop eating it. The agent hype cycle arrived later and mostly made that harder to sell honestly.
So we work the other way round. We start with the process map and a baseline, decide whether a model is even the right tool, and build the smallest system that removes the manual steps. Where the work has financial or legal consequences, a person stays in the loop — by design, with a threshold, not a paragraph in a contract.
Everything we build is written in your repository, deployed in your cloud if you prefer, and handed over with the evaluation set that proves it works. If we disappeared tomorrow, your team could keep running it — that is the test we hold ourselves to.
Evidence over enthusiasm
Every claim about quality comes from an evaluation run on your data. If we cannot measure it, we do not assert it.
Humans where it matters
Automation takes the repetitive work. Decisions with legal, financial or safety consequences keep a person in the loop.
You own the result
Code, prompts, evaluation sets and documentation are yours. No proprietary layer that only we can operate.
Discovery is not a workshop with sticky notes. It is hours with the people who do the work, watching where the time actually goes.
How the team is shaped
-
Delivery lead
Owns the process map, the plan and the weekly cadence. Your single point of contact.
-
AI engineer
Models, prompts, retrieval, evaluation harness. Writes the code that decides things.
-
Integration engineer
APIs, ERP/CRM adapters, durable workflows, failure handling, monitoring.
-
Evaluation & QA
Builds the golden set, runs regressions, measures accuracy, cost and drift.
Tooling
We are opinionated
about tools, not loyal to vendors.
Choices are made on your data during the pilot and documented, so a future engineer — yours — can see why.
Models
- Frontier APIs where quality leads
- Open-weight models self-hosted in your VPC
- Classical ML and rules where a model is the wrong tool
Orchestration
- Durable workflow engines with idempotent steps
- Queueing, retries, timeouts, compensation
- Event-driven triggers from your systems
Data
- Postgres for state and audit
- Vector + keyword retrieval with citations
- Warehouse and lakehouse integration
Interfaces
- Review consoles your operators can live in
- Slack, Teams and email surfaces
- APIs and webhooks for your own product
Infrastructure
- Your cloud account or ours
- Infrastructure as code, reproducible deploys
- Metrics, traces and cost dashboards
Next step
Work with a studio that leaves.
We are hired to make ourselves unnecessary: your team runs it, your repository holds it, your evaluation set proves it.
One business day — and the reply comes from an engineer, not a sales sequence