QuantHall is a SaaS platform that combines machine learning ensembles, explainable AI and genetic agent systems to turn complex quantitative datasets into transparent, traceable, decision-grade insights.
Leave us your details and we will get back to you.
What QuantHall does
Machine learning, explainability and genetic agents for quantitative data.
QuantHall is an AI software platform for the quantitative analysis of complex datasets. It combines ML ensembles for predictive scoring, SHAP for transparent attribution and adaptive agent systems for decision support — engineered to keep every output traceable, reproducible and explainable.
Disclaimer
QuantHall provides software and analytical outputs. Results are technical computations and should not be interpreted as professional advice, recommendations or guidance of any kind.
Platform
QuantHall harmonises heterogeneous quantitative datasets into a single analytical layer — structured records, time-series, categorical groupings and auxiliary signals — ready for modelling, evaluation and attribution.
The ML ensemble combines LightGBM and XGBoost with 55 engineered features and 22 specialised sub-models. Each prediction is decomposed feature-by-feature with SHAP, so every output is fully traceable.
Walk-forward evaluation eliminates look-ahead bias and reports stable out-of-sample behaviour across quarterly windows. Every metric is reproducible from the same code, on the same data.
Capabilities
Each module is a building block of the platform — covering data ingestion, modelling, explainability, evaluation and adaptive decision support.
522 agent variants organised around 13 archetypal styles, evolved across 30 generations under a multi-objective fitness function. Produces adaptive, weakly-correlated decision signals from heterogeneous inputs.
Every model output is decomposed into SHAP contributions across all 55 features. The platform exposes why a number is produced — not just what number — so every result is auditable end-to-end.
A separate model is trained per evaluation window to eliminate look-ahead bias. The resulting metrics reflect genuine out-of-sample behaviour rather than in-sample fit.
A 0–100 analytical score per entity, produced by an ensemble of LightGBM and XGBoost with 55 features and 22 specialised sub-models. Updated continuously as new data arrives.
Reproducible projections under three scenarios (low / base / high), calibrated with category-specific overrides. Useful for sensitivity analysis and what-if studies.
Ingestion, cleaning and harmonisation of heterogeneous sources into a single analytical schema with 18 dimensions of structured, time-series and categorical signals.
Allocation studies under category-level concentration constraints, with attribution by quality quadrant and quintile bands. Suitable for any constrained selection problem over scored entities.
Comparative analytics that adapt the relevant metric set to each category, so entities are compared on the dimensions that actually matter inside their cohort.
The same modelling approach applied to time-series of aggregates and cross-sectional snapshots — including benchmark comparisons and inter-class differential analysis.
External numerical signals — auxiliary indicators, contextual time-series, categorical metadata — are integrated as context features alongside the core dataset, enriching every prediction.
Decomposition of larger populations into cohorts, with time-series of how model output distributes across each segment over evaluation windows.
About QuantHall
QuantHall is a SaaS platform that combines machine learning, explainability and genetic agent systems to analyse complex quantitative datasets — turning raw data into transparent, traceable, decision-grade output.
The platform is engineered around three principles: reproducibility (every result reconstructible from the same code and data), explainability (every output decomposed into its drivers) and out-of-sample evaluation (no look-ahead, no in-sample illusions).
QuantHall SL is built on two complementary segments: rigorous quantitative AI and privacy-preserving data engineering. The second is BlindLink, our anonymisation and pseudonymisation line for sensitive clinical data — a key and fast-growing part of the company, detailed below.
QuantHall provides software and analytical outputs. Results are technical computations and should not be interpreted as professional advice of any kind.
Features, models and evaluation protocols are documented end-to-end. Every output can be traced back to its inputs.
522 agent variants evolved across 30 generations under a multi-objective fitness function — producing adaptive, weakly-correlated decision signals.
Walk-forward validation across every evaluation window. Reported metrics reflect out-of-sample behaviour, not in-sample fit.
LightGBM + XGBoost ensemble with SHAP attribution. Every prediction is decomposed feature-by-feature so it can be audited end-to-end.
Anonymisation and pseudonymisation of sensitive clinical data so it can be shared and used with AI. Data protection is a core company segment.
A core QuantHall SL segment · Health data privacy
BlindLink anonymises and pseudonymises Spanish clinical and administrative text so hospitals, insurers, foundations and public administrations can share it and use it with AI — without the original data ever leaving their perimeter. Each entity is handled with one of four strategies (redact, mask, reversible pseudonym or synthetic value) in under 30 seconds per document.
AES-256-GCM envelope encryption, separated keys and an immutable HMAC-signed audit log. Detection models are self-hosted — no data is sent to third parties to anonymise.
Preserves structure and statistical utility, so anonymised data stays valuable for research cohorts, analytics and for training or evaluating AI models.
A pseudonym vault gated by MFA and explicit roles allows controlled re-identification for clinical use, with every action traced end-to-end.
Four strategies, chosen per entity
The value is removed and tagged ([NAME], [DATE]). For open scientific publication.
The value is hidden behind *** while keeping the document readable. For demos, training and screenshots.
Replaced by a reversible HMAC token. For research cohorts that may need controlled re-identification later.
Replaced by a realistic synthetic value (e.g. a plausible fake name). To train AI while preserving the statistical distribution.
In the health sector this powers research and CRO data sharing, clinical-AI training and evaluation, insurer claims analytics, and coding or registry submissions — all without exposing personal data.
Designed to align with RGPD/LOPDGDD, the EU AI Act, EHDS and the ENS High category — built on a Spanish clinical NLP stack and deployable on-premise or in an EU cloud region.
Contact
For information about QuantHall and its capabilities, please reach out by email and we will get back to you.
info@quanthall.comWe typically reply within 1–2 business days.