Research timeline
1. FY2025
The year started where 2024 left off. Feedback from the rail sector on the September 2024 demo had accumulated, and the first three months of 2025 went into working through it — adjustments to the analytical assumptions, to how results were presented, to what the demo could and could not credibly claim. The pattern that emerged from those exchanges then fed the grant proposals and applications written through the spring; those documents were where the theoretical basis of what we would ship in the autumn was first laid out in writing.
The real platform came together between August and November. We refer to it internally as our "factory": the data and compute pipeline that turns raw geospatial feeds into the per-asset and portfolio metrics that customers actually consume. The first end-to-end production run was the level-crossing characterization, delivered to SNCF Réseau in November 2025 and validated against field data in December. The model covers every active level crossing on the French national network across roughly 200 geospatial dimensions, drawing on the CEREMA inspection grid for category-2/3 crossings and the EPSF technical referential. The output is a portfolio view that supports filtering across the whole network and a per-crossing view that exposes origin–destination pair metrics.
1.1 External activity
Across the year, twelve industry conferences and trade shows, ten prospection meetings or product demonstrations, and six financing dossiers or letters of support.
2. FY2024
The first full year of Oriskami SAS was about choosing — choosing the data sources to build on, the analytical approach to commit to, the engineering substrate to lay down. Roughly nine months of prototyping went into those choices, and they converged in a first demonstrator shown in September 2024. That demo was extra muros by design: it ran disconnected from any live data feed, built to convey the analytical logic before pretending to be production. We took it to the rail sector to gather the feedback that would shape the next year's work.
2.1 Platform foundation
The architectural blueprint that came out of this period — how raw data turns into results — is described on the home page. What FY2024 settled was the substrate underneath that chain: data ingestion from OpenStreetMap, accidentology datasets, GTFS feeds, and remote-sensing layers from NASA and ESA, on top of an operational stack tuned for repeat-runs of the same analytical pipeline.
2.2 Component maturity tracking
Knowing which parts of the codebase were still research and which were on the path to industrialization mattered both internally and for our R&D filings. We adopted a TRL-style maturity classification (Readiness Levels) and tracked each component against it — a frame that lets us see, at any moment, what is still being explored, what is being hardened, and what is already in industrial use.
2.3 Identified challenges
Four areas were flagged for follow-up work: user experience and workflow completeness, scalability of the compute infrastructure, robustness of the AI components under distribution shift, and interpretability of model predictions.
3. CIR positions
Two distinct CIR positions sit alongside the year-by-year work, each with its own scope and recognition. They are described separately below so each can be read on its own merits.
The first is the company's. As a software editor, Oriskami SAS files its own annual CIR (Crédit d'Impôt Recherche) on the R&D it carries out on SAMRoute — the operations described in the year sections above are what those filings document.
The second is the founder's. Independently of the company's filings, Fabrice Colas holds a personal CIR-approved consultant agrément issued by the French Ministry of Research, first granted in 2022 and renewed in May 2025 for the 2025–2027 cycle (with a parallel CII — Crédit d'Impôt Innovation — agrément for the same period). This agrément is what authorizes other companies to engage him for R&D consulting work that is itself eligible for the CIR on their side.
Read side by side, the two positions describe two complementary credentials. One places the company inside the French R&D framework as a software editor with its own research programme; the other places the founder, individually, on the public list of consultants the State authorizes other companies to engage for R&D activities. Both stand on their own merit, and together they situate the work within the recognized research ecosystem in France.
4. Ecosystem
Across 2024 and 2025 the work has been supported by OVHCloud (Startup Program credits, February 2024 – February 2025), CEREMA (advisory, the public infrastructure-engineering reference in France), SNCF Réseau (pre-sales discussions on level-crossing evaluation, leading to the first production delivery in November 2025), and the DGFiP (which processes the company's annual CIR refund).
