An independent research institute

Recovering latent structures from observed data.

Senior researchers working on inverse problems through two neuro-symbolic methods — consolidated SciML, and LLM-orchestrated RAG-mediated calibration of complex simulators.

Founded 2026 · Cambridge UK + Brussels BE · No physical premises · Distributed across Europe
· · ·
I.

The inversion act.

The Institute reads the inverse problem as a precise operational act: a material to be inverted (heterogeneous evidence) is converted into a configuration accepted by a functional receiver (a structured model). The roles are not fixed — they materialise only at the moment of inversion.

Line B · consolidated SciML

Data + system knowledge → physical model

Measurements and structural priors inverted into parameters and closures for ODE/PDE/state-space models, augmented by PINNs, UDEs, or graph neural networks.

Line A · frontier method

RAG + LLM → process model

A hybrid syntactic-vector RAG under LLM orchestration assembles the parametric profiles and boundary structures a mechanistic simulator accepts.

Read the full Founding Manifest →

II.

Two lines, six sub-areas.

Line A Frontier · RAG-LLM

  • Foundation TRAG — hybrid RAG on the EMA corpus (operational). trag.argentesi.com
  • A·1 Lake Maggiore Daphnia — RAG-fed compartmental ODE inversion (M. Manca).
  • A·2 Lake Varese — RAG-calibrated GLM-AED2 biogeochemistry.

Line B Consolidated · SciML

  • B·1 Cladocera — McKendrick–Von Foerster transport + PINN birth-rate inversion.
  • B·2 Material accountancy — NUMSAS / loss-pattern as state-space neural inversion.
  • B·3 Radioecology Zn-65 — graph neural network for tracer kinetics (Merlini 1971).

Project portfolio in detail →

III.

People & archive.

The Library holds five decades of founding-nucleus work (1971–2026) — scanned originals with English regenerated editions, filterable by line and recovery status.

Full nucleus →    Browse the catalogue →