Torvo Construction

Inspiration

I am not a trained software engineer. I am a 28-year automotive diagnostic technician from Rugby in the UK, and Torvo started from a simple frustration I have seen across trades, care, construction, garages, clinics and public services: skilled people do the real work, then lose hours proving they did it.

In less than two months, with GPT, Claude and Codex as the build layer, I went from zero to a working multi-vertical platform. Torvo's core idea is deliberately simple:

1. Listen to the real-world work.
2. Understand what actually happened.
3. Generate the correct professional document.
4. Let the human review, correct and approve it.

The tagline is: the paperwork does itself.

Construction is the strongest proof because the admin pain is extreme and the stakes are real. A site walkthrough should not vanish into memory, WhatsApp, paper notes or a blank RAMS template. It should become an evidence pack that a supervisor can review, improve and sign.

What it does

Torvo Construction turns a spoken site walkthrough into a reviewable construction evidence pack.

A user records on a phone while walking the job. Torvo captures the transcript, site context, hazards, equipment, access issues, stored energies, materials, weather, people, controls and follow-up actions. It then drafts the documents the team would normally have to create manually:

• Risk assessment
• Energies assessment
• Site diary
• Toolbox talk
• RAMS / method-statement style content
• Permit-to-work checks
• Asbestos consideration prompts
• Equipment inspection and certification prompts
• Materials and stored-energy notes
• HV/EHV advisory banners where high-voltage context is detected

Torvo does not pretend to be the competent person. It drafts, flags and explains. The human remains responsible, edits the pack, checks source evidence, and confirms before anything is treated as approved.

Construction is the spearpoint, but the capability is wider. The same platform pattern is already being explored across Field Service, Treatment, Hospitality, Offshore, Police/BADGE, Community Care, SEND education and AutoDiag/workshop operations. The common problem is not construction paperwork alone. It is professional proof-of-work.

How it works

The construction app is a phone-first PWA with a field recorder and a desktop review workspace.

The workflow is:

1. Record the walkthrough on site.
2. Transcribe the real speech.
3. Extract structured site facts, hazards, controls and evidence.
4. Generate the document pack.
5. Surface Niro advisory notes with fact, consequence and recommendation.
6. Let the user edit, correct, dismiss, restore and version documents.
7. Keep an audit trail of review, sign-off and safety prompts.

The live build has gone beyond a static demo. It includes document modal editing, version history, GPS site context, mobile stale-state recovery, HV/EHV advisory detection, construction skill packs, and public sample output documents.

How I used Codex and GPT-5.6

Torvo existed before OpenAI Build Week, so I am being explicit about what is new and how Codex/GPT-5.6 was used.

The bigger story is that I built Torvo as a non-coder by working with AI systems as my engineering team. GPT and Claude helped me reason, learn and build. Codex helped turn that into working code, tests, server changes, browser checks and production-quality workflows.

For Build Week, Codex with GPT-5.6 was used to:

• Re-read the Torvo memory, server context and current product state.
• Select construction as the strongest vertical rather than submitting a weaker broad pitch.
• Audit the Devpost requirements against the live project.
• Reframe Torvo as one platform with construction as the proof vertical.
• Prepare the submission copy, testing notes and demo-video structure.
• Validate public proof assets and the judge path.
• Document what is pre-existing, what was extended, and where Codex/GPT-5.6 contributed.

Codex was not just used to generate snippets. It acted as an operating partner: reading context, checking live URLs, inspecting forms, editing content, testing assumptions, and helping a non-traditional founder package a real working system for judging.

Challenges

The hard part is trust.

In construction, a system cannot sound like it is issuing instructions it has no authority to issue. It cannot silently invent controls, miss high-voltage context, leak one organization's data into another, or make the UI look finished when the legal/audit state is not.

That forced several product decisions:

• Human confirmation before approval.
• Advisory language, not command language.
• Build and field surfaces separated.
• Audit logs for safety prompts.
• Clear source evidence behind generated documents.
• Mobile recovery when processing state changes on another device.
• High-voltage detection that avoids false positives for ordinary low-voltage references.

What I am proud of

I am proud that this is not a slide deck.

Torvo has real verticals, real services, real routes, real generated documents and a real field-cohort path. The construction proof includes a first real walkthrough, public sample PDFs, and a product architecture that can transfer into other high-admin professions.

I am also proud of the personal leap. Less than two months ago I was not a software developer. With AI as the build layer, I have been able to create something that would normally require a team: product design, backend code, phone capture, document generation, audit trails, UI, server deployment, and multi-vertical thinking. Two UK patent applications have been filed around Torvo's capability.

What is next

Next for Torvo Construction is a judge-safe sandbox, a stronger public demo route, and a continued push from construction proof-of-work into field service, clinics, clubs, offshore operations and other professional domains.

Long term, Torvo is not trying to replace experts. It is trying to give experts their time back.

The paperwork does itself. The human stays in charge.

Built With

• claude
• codex
• construction-technology
• gpt-5.6
• human-in-the-loop
• openai
• progressive-web-app
• voice-ai