Tempworks.ai
Co-founder · Product, AWS architecture, design and delivery · Exploring seed funding · 2026
I took Tempworks AI from an idea to a production, multi-tenant SaaS platform for UK temporary works firms. It brings jobs, documents, engineering workflows and company knowledge into one secure conversational workspace.
I owned the product direction, AWS architecture, application build, UX, design system and brand. My co-founder supplied the engineering domain expertise; I turned it into software that could automate real work without removing human accountability.
The defining constraint was safety. The system can extract a brief, run a bounded workflow and produce drawings and calculation packs, but it cannot invent missing engineering values or bypass a competent person's review.
This was my first time taking a product of this technical scope from concept to production. Over an intense nine months, I learned unfamiliar systems while building them, using focused research, written decisions, small reversible releases and test gates to keep mistakes contained and meet delivery deadlines.
What I built
- AI product platformA conversational interface combining chat, connectors, documents, engineering workflows and generative dashboards with the CRM system of record
- API bridgeA secure application layer connecting the AI experience to jobs, projects, revisions, documents and company data in the CRM backend
- Product MCPA user-scoped integration that lets external chat surfaces such as WhatsApp and Telegram access Tempworks safely without a shared privileged account
- PiersOur internal company agent: a persistent company brain and business partner that retains operational context, coordinates work and connects the team to Tempworks tools
- BrandPositioning, identity and product narrative for a new category of AI-native engineering software
- Landing pageA product-led website that makes the platform, its safeguards and its commercial value clear to non-technical buyers
- App design systemA shared system of tokens, components and interaction patterns connecting the product experience to the Tempworks brand
- Engineering workflow systemThe product's proprietary core: recordings of engineers completing real jobs converted into reusable agent skills and AutoCAD plugins that run on EC2, producing checked drawings and calculation packs
Key decisions
- Enforce tenancy and RBACCognito identifies the user; company membership and role capabilities are verified on every request, with PostgreSQL row-level security as a second boundary
- Constrain the AI boundaryBedrock has no database access; tools use the signed-in user's permissions, retrieved evidence is validated, and engineering outputs require human review
- Delegate without impersonatingTelegram and MCP access is linked to a real user, with short-lived tokens and an entitlement check each time credentials are issued
- Keep customer data privateAPI Gateway and a VPC Lambda are the only application path to private RDS; files use controlled, version-bound S3 access
- Make CAD automation fail safeImmutable templates, capacity checks, quarantined partial files and two zero-failure verification passes before release
- Engineer for UK GDPRUK data residency, scoped access, MFA, audit logs, retention controls, reviewed exports, journalled erasure and recovery drills
- Design for recoveryIsolated staging, 68 live acceptance checks, policy-as-code and rollback-first handling of a production schema incompatibility
Outcome
Within nine months, Tempworks AI progressed from concept to a production platform and was piloted with two UK engineering firms. It proved that specialist engineering work could be automated end to end while preserving traceability, tenant privacy and human sign-off.
Architecture
AWS in London: ECS Fargate, Lambda, API Gateway, RDS PostgreSQL, Cognito, S3, Bedrock, EventBridge and CloudWatch. Infrastructure is defined with CloudFormation and checked with policy as code. Delivery runs through tested, reviewable releases rather than manual production changes.