Evidence · Flagship

Helionics: vertical AI that turns an industrial inbox into verifiable records

Founder proof. Not a slide about agentic AI. A company that ships it, with a manufacturer module live in production.

Role

CEO & Founder

Sector

Industrial / Oil & Gas supply chains

Deployment

60 days, one workflow, one engineer

Status

Manufacturer module live in production

The problem

Capital projects in oil and gas run on paper, email and PDFs. Everyone has an ERP, a PLM and an engineering stack, and none of them talk to each other. So the record book is built by hand at the end, cash sits in finished work for months, and because none of it is verifiable, a certificate can be reused, a test result nudged, a zero changed in a spreadsheet.

The answer is already in the inbox. One project can mean a hundred thousand emails. Search for the word and you drown; search for the fact and you miss it. The industry failed to solve this for twenty years. It is solvable now because the AI is finally good enough.

What we built

Vertical AI, industry-trained. Not a wrapper on a general model. Helios reads the purchase orders, material certificates, inspection plans and deviation emails a supply chain already produces, extracts the structure, and ties every certificate to its line item the moment it lands, bound to the physical asset by heat number, tag number and serial.

Email mining, proven at scale. AI agents read messy supplier emails and PDF attachments, scanned records, handwritten sign-offs, highlighted lines, and pull out the facts, even the ones that never name themselves. Proven on a public corpus of 500,000 real emails. A thousand people could not do this by hand; it is a pattern problem, not a volume problem.

The intervention board. A dashboard shows data. An intervention board tells the right person what is broken, why it costs money today, and exactly what to do about it. Every action is ranked by schedule impact and cost of delay, each with an owner and a decision. You see the delay the day it happens, not the month after.

Human-in-the-loop by design. The AI surfaces the next action while it still matters, and a human accepts it. Recommendations, not autopilot. Drafted responses go out only after human sign-off. The decision stays with your people; the evidence stays with the record.

EventChain. The verification layer underneath: an open-source, append-only, cryptographically anchored protocol. Sensitive payloads stay encrypted and access-controlled; the event log carries only hashes. Every record is tamper-evident, time-anchored and independently verifiable, without blockchain's scalability problems.

Why it matters

Most AI consultants advise on systems they have never had to ship. Helionics is my own P&L: I designed the agent architecture, the human-in-the-loop governance, the 60-day deployment method. I carry the consequences when it does not work. That is the standard I bring to client engagements.

500k

Emails mined, public-corpus proof

60 days

Kickoff to verifiable production record

×35

Faster record-to-invoice, pilot-evidenced

Open

EventChain protocol, open source