Stathmon

Custom AI systems, built around the way your company works.

Stathmon

We design and build AI systems for pharmaceutical, energy and finance companies, as well as research groups. Each is shaped around the people who will use it, and can run on your own infrastructure so your data stays in-house.

See the work
20+peer-reviewed scientific publications
5+corporate and academic clients
Athensbased
Selected clients
WinMedicapharmaceutical VisionITbanking software Hellenic Passive House Instituteenergy & buildings Lab.RACSE, NTUAenergy research VP&Cpharmaceutical policy
Work

Systems in daily use, and the research behind them.

01 PharmaceuticalWinMedica
Fig. 1Approved text aligned with the printed artwork Approved text Printed artwork 1 difference found

Labeling and artwork comparison

Every word on the pack checked against the approved text. Leaflets, cartons and blisters, aligned word by word, every difference marked where it is. In daily use by three departments.

The LabArt page
02 Pharmaceutical policyVP&C
Fig. 2A digest built around the client's interests
Intelligence digestThursday 16 October
3 items, all verified
Pharma regulatoryPricing & reimbursementTendersClinical
Variation guideline updated for labelling changes
New wording on when a label change needs a prior approval.
Agency guideline, 15 Octverified against source
October price list published, two products in the lower band
Effective from 1 November; the previous list stays valid until then.
Committee bulletin, 16 Octverified against source
Hospital tender for oncology supplies closes 30 October
Specification unchanged from last year's call.
Tender notice, 14 Octverified against source

Personalised intelligence gathering

A personalised intelligence gathering system. A digest built around the client's interests, pharma regulatory, pricing and tenders, with every item checked against its source before it reaches the reader. Used by large pharmaceutical companies and policy-makers.

03 EnergyWinMedica
Fig. 3Site, plant and grid Solar array1,553 kWh per kWp a year Batterycharged cheap Pharmaceutical plantits own load, forecast Grid 48-hour site weather forecast buy / sell

Solar and battery sizing for a factory

End-to-end analysis of a solar and battery plant for a pharmaceutical factory. Site weather, market price and plant load forecasting, energy yield, battery dispatch, financial feasibility and lifecycle analysis, using methods from our published work. Energy efficiency, reliability and revenue, each quantified.

04 EnergyLab.RACSE, NTUA
Fig. 4Revenue over a year, against the theoretical maximum
theoretical maximumour controllerfixed rule

Battery trading controller

82% of the theoretical maximum. The maximum is what a battery would earn if it knew tomorrow's prices and had the perfect plan. A learned controller reaches 82% of that with real forecasts, and outperforms all the popular industry methodologies.

05 EnergyLab.RACSE, NTUA
Fig. 5Greek day-ahead market, one day
forecast, made the day beforecleared price

Electricity price and weather forecasting

Tomorrow's prices on the Greek day-ahead market, and the weather that drives them. Deep-learning forecasters for hourly prices and for 48-hour temperature, irradiance and humidity, published and scored on a full year the models never saw.

06 FinanceA private investor
Fig. 6Thirty years of historical simulation
the portfolioa single market indexcrisis periods
Risk toleranceLowMediumHigh

Automated portfolio

An automated, diversified portfolio with a risk level you choose. Built on published allocation methods and tested on official multi-decade data, through the major market crises in that period.

07 In-house productStathmon
Fig. 7The app, on your own server
Right Hand
07:40Brief sent. One push held overnight, delivered.
Morning. Three things fit today: the Acme quote (2h), the server deploy (half a day), the Meridian follow-up (20m). Maria at Acme expects the quote by Friday.
Move the Acme quote to Friday and remind me Thursday morning.
Done. Quote due Friday, reminder Thursday 08:00. Nothing else changed.
Send revised quotemoved2hFri
ChatTasksMore

Right Hand, a personal assistant

An assistant that gets to know you. Plain-language chat that learns how you work, stays out of the way, and reports exactly what it changed. Tested for reliability, stress-tested for security, and able to run entirely on your own server. App and web.

The Right Hand page
Also built
08

Custom word segmentation pipelines

Local detectors that find every word on pharmaceutical artwork, trained on the client's own documents. More exact than the strongest commercial alternative on the same pages.

WinMedica
09

Pharmaceutical OCR, Greek and English

Reads Greek and English pack text more accurately than the commercial cloud OCR services, because it is trained on the client's own documents.

WinMedica
10

Greek invoice reader

Structured fields from scanned invoices of varied layouts, including the myDATA MARK. Validated and commercialised.

VisionIT
11

Private language models

Open-weight models deployed and benchmarked on your own GPUs or HPC allocation, so data stays on-premise.

12

Heat-pump performance across refrigerants

Regression models that predict cycle performance, including fluids not seen in training.

Lab.RACSE, NTUA
13

Plugins for energy design software

Customised plugins for popular building energy software such as IDA ICE, running automated design optimisation on the model instead of trial and error.

Lab.RACSE, NTUA
Who builds it
Georgios Vamvouras

Georgios Vamvouras

Founder and principal engineer. Dipl.-Ing. Mechanical Engineering, National Technical University of Athens.
0120+ peer-reviewed scientific publications
02Trusted by large and small companies and by university research groups
Personal site and publications
ContactAthens
[email protected]

Tell us the workflow, data or decision you want to improve.