Questions about
LabGPT

LabGPT produces audit-grade environmental field data for organisations with a reporting obligation. It plans a sampling campaign in software, collects samples at your sites with autonomous drones, has them analysed at ISO/IEC 17025-accredited partner laboratories, and returns each result mapped to the ESRS, TNFD, GRI or SBTN datapoint it answers. Every result carries the GPS coordinates, timestamp and unbroken chain of custody it was collected under, written by the system at the moment of collection.

Environmental reporting is now mandatory across Europe and regulators fine organisations that do not comply, but the data behind those reports is still collected by hand, written down by hand, and carried to a laboratory by hand. In contaminated soil, more than 90% of the error in a result comes from how the sample was taken rather than how it was analysed, and one study measured sampling error at over 20 times the analytical error. The uncertainty is in the field, not the laboratory — so that is where LabGPT works.

The buyer is the person accountable for an environmental filing and answerable if the data behind it cannot be defended: an EHS manager, a sustainability lead, or the environmental consultancy acting for them. Typical users are organisations reporting under CSRD/ESRS, companies managing contaminated or industrial land, and those setting nature and biodiversity targets that need measured site data rather than desktop estimates.

No. All analysis runs at ISO/IEC 17025-accredited partner laboratories. LabGPT designs the campaign, collects the sample, maintains the chain of custody, and returns the result with the method that produced it. You can approve the specific laboratory before any sample is sent, and if you would rather use your own laboratory, LabGPT collects and delivers to it.

It means an assurance provider can trace a number back to a method and a measurement. Three things make that possible: the same documented protocol is followed every round and filed with the result, so nothing silently changed between rounds; every result carries the coordinates, timestamp and custody chain it was collected under; and the data is returned at two levels — headline indicators for the filing itself, and the full method-level record underneath for a regulator who asks for one.

LabGPT was founded by Alice Chen (CEO) and Rohan Talwadia (CTO), and works with advisors in environmental science and technology systems. See the team page.

Primarily the EU sustainability reporting regime and the soil rules that sit alongside it: ESRS E2, E3 and E4 under CSRD, and the EU Soil Monitoring Directive. Results are also returned in the form the Science Based Targets Network requires for location-based target setting, mapped to TNFD and GRI disclosures, and as the primary site data a natural capital account needs. The obligation is no longer to report. It is to evidence.

They require pollution, water and biodiversity to be reported for each individual site, not as a company-wide average — which is what makes site-level measurement unavoidable. In Germany, non-compliance has carried fines of up to €10 million or 5% of group turnover, plus director liability, since January 2026.

It reduces how many datapoints you report, not the standard of evidence behind the ones that remain. The European Commission adopted the revised standards in July 2026, cutting mandatory datapoints by over 60%. Fewer datapoints does not mean weaker evidence — the ones that survive still have to be supported.

It came into force in December 2025 and requires soil health to be monitored against a common list of descriptors: organic carbon, erosion, compaction, contamination and soil biodiversity. It must be written into national law by December 2028. A PFAS contaminant list follows within 18 months of entry into force, and a public register of contaminated sites within ten years.

Environmental DNA is genetic material organisms shed into water, soil and sediment. Sampling it establishes which species are present at a site without needing to observe or capture anything, which makes it well suited to the site-level biodiversity evidence ESRS E4 and TNFD ask for. In a LabGPT programme, the drone collects the sample and an accredited partner laboratory runs the analysis; results come back as species presence, microbial community profile, pathogen screening, and invasive or introduced-organism detection.

The system writes the GPS coordinates, the timestamp and each custody transfer at the moment of collection, not afterwards from field notes. That record travels with the sample to the laboratory and is returned attached to the result, alongside the method that produced it. This is the part manual sampling most often cannot reconstruct after the fact.

That is what the record is designed for. Assurance providers test whether a number traces back to a method and a measurement, so the protocol is documented, filed with each result and repeated identically every round; each result carries its own coordinates, timestamp and custody chain; and the full method-level record sits behind the headline figures if a detailed audit is requested.

Airspace and ground-constraint validation is part of the planning step, before any flight is designed. Requirements differ by site, sector and member state, so they are scoped rather than assumed: during scoping LabGPT confirms in writing what is already in place, what is obtained for your site, and what falls outside scope. A site-specific safety case is written before any flight is planned.

You own it outright — you can export it, share it, or sell it. It is held on EU-based servers and is GDPR compliant.

A programme is priced on three things: how many sites, how many rounds a year, and which programme. Two components are priced separately — a service fee covering collection, the platform and reporting, and laboratory analysis, which is passed through per sample. Parameters measured in flight carry no laboratory cost. Duration is agreed up front; most clients run a baseline once and then a recurring programme measured against it. Final pricing is confirmed on scoping, because it varies with the number of sites, the area of each, coverage and the depth of data.

A consultancy sends people to site. The laboratory bill is the same either way; what changes is that none of your staff spend field days on site, every round also returns site-wide mapping and sensing that manual sampling does not collect, and the sampling protocol is executed identically every round — so a change in the data reflects the site rather than which contractor turned up.

Book a 30-minute call. LabGPT will design and simulate a sampling mission for one of your sites, present it for your approval, and scope what a programme would cost. Starting with a single site is normal and is usually a Site Baseline — one round, establishing the reference record everything later is measured against.