# Özer Cem KELAHMET > Özer Cem KELAHMET builds Data Language Models, large tabular foundation models, and real-time AI systems. This site holds the documentation of the large tabular foundation model (Schema-2), articles on tabular AI, and projects. Preferred name: Özer Cem KELAHMET. Contact: ozer@ozercemkelahmet.com. LinkedIn: https://www.linkedin.com/in/ozercemkelahmet. Paper: https://arxiv.org/abs/2605.06290. Every page has a Markdown twin at the same URL with `.md` appended. The full corpus is at https://www.ozercemkelahmet.com/llms-full.txt. ## Key facts - **What is a large tabular foundation model?** A large transformer pre-trained on a very wide range of tables, so it can predict, fill in missing values, flag anomalies, and match records on a new table in one pass, without training a new model for each dataset. - **What is a Data Language Model (DLM)?** A Data Language Model is a foundation model that reads raw tables directly, the way a language model reads text, with no serialization, no preprocessing, and no column names. The category is defined in the arXiv paper 2605.06290 by Özer Cem KELAHMET. - **What is Schema-2?** Schema-2 is the large tabular foundation model built by Özer Cem KELAHMET. It has more than 1.5 billion parameters, uses axial attention over rows and columns, a Perceiver-style global path, and a schema path, and one checkpoint of about 2.18 billion parameters carries every capability. - **What can the model do?** Row-level prediction, missing-value imputation, anomaly detection, probabilistic regression, continual learning across tasks, entity and schema matching, blind sector identification, and prediction that does not depend on column names. - **Who is Özer Cem KELAHMET?** A technology and AI leader who builds Data Language Models, large tabular foundation models, and real-time AI systems, end to end from research and architecture to production. - **Where can I read the paper?** The paper is on arXiv: https://arxiv.org/abs/2605.06290. The documentation on this site covers the architecture, pre-training, evaluation, and every benchmark. ## How to cite KELAHMET, Ö. C. Data Language Models: A New Foundation Model Class for Tabular Data. arXiv:2605.06290. https://arxiv.org/abs/2605.06290 ## Large tabular foundation model documentation - [What it is](https://www.ozercemkelahmet.com/model/model-overview.md): What a tabular foundation model is, what it reads, what it returns, and how it differs from training a model per table. - [Research](https://www.ozercemkelahmet.com/model/research.md): The paper behind the model: why tabular data has no native foundation model, and what a Data Language Model is. - [Solutions](https://www.ozercemkelahmet.com/model/solutions.md): One foundation model across every industry. Your data ships as tables: exports, extracts, feeds, logs. The model reads them raw and answers six questions that every team running on tables ends up asking. - [Tabular foundation models](https://www.ozercemkelahmet.com/model/tabular-foundation-models.md): What a tabular foundation model is, the landscape of approaches from boosted trees to data language models, and why reading the raw table matters. - [Architecture](https://www.ozercemkelahmet.com/model/architecture.md): How a tabular foundation model is built: cell encoding, axial attention, Perceiver-style global reasoning, task heads, adapters, and one merged checkpoint. - [Cell encoding](https://www.ozercemkelahmet.com/model/cell-encoding.md): How a tabular foundation model turns each raw cell into a vector: value, type, position and explicit missingness, with column identity zeroed. - [Pre-training](https://www.ozercemkelahmet.com/model/pretraining.md): How a tabular foundation model learns before it sees your table: masked cell modeling, synthetic data priors, distributed training, and checks. - [In-context prediction](https://www.ozercemkelahmet.com/model/in-context-prediction.md): How a tabular foundation model predicts without training on your table: labelled rows act as a support set, and new rows are read off by comparison. - [Continual learning for tables: learning new tasks without forgetting old ones](https://www.ozercemkelahmet.com/model/continual-learning-for-tables.md): How a tabular foundation model learns task after task without forgetting: adapters, replay of 48 raw rows per task, what failed, and what stays open. - [Column name independence: how a tabular model works without headers](https://www.ozercemkelahmet.com/model/column-name-independence.md): - [Compression and serving: running a 2.18 billion parameter tabular model](https://www.ozercemkelahmet.com/model/compression-and-serving.md): How a tabular foundation model is compressed to half, 8-bit and 4-bit precision, how fidelity is checked, and what changes when it serves requests. - [Capabilities](https://www.ozercemkelahmet.com/model/capabilities.md): What a tabular foundation model returns in one pass: sector, column profile, cross-table map, imputation, predictions, and a held-out score. - [Evaluation](https://www.ozercemkelahmet.com/model/evaluation.md): How a tabular foundation model is evaluated: held-out splits, benchmark protocols, sealed configurations, contamination checks, and what a score claims. - [Limitations](https://www.ozercemkelahmet.com/model/limitations.md): What a tabular foundation model does not do: data limits, task limits, benchmark caveats and open problems, so results are read correctly. - [Benchmarks](https://www.ozercemkelahmet.com/model/benchmarks.md): What each benchmark measures, how I built the capability that answers it, and what made the difference, in the order the work was done. - [Row-level prediction benchmark: how the tabular foundation model was built and measured](https://www.ozercemkelahmet.com/model/bench-row-prediction.md): How a tabular foundation model was built and measured on row-level prediction boards: classification and regression, ranked by Elo against tuned pipelines. - [Continual learning benchmark: keeping old tasks while learning new ones](https://www.ozercemkelahmet.com/model/bench-continual-learning.md): How a tabular foundation model learns task after task without forgetting: the failures, the replay recipe that stayed, and what stayed open. - [Anomaly detection benchmark: finding rows that do not fit, with no labels](https://www.ozercemkelahmet.com/model/bench-anomaly-detection.md): How a tabular foundation model ranks anomalous rows in the semi-supervised ADBench setting, and how label-free channel selection made the difference. - [Probabilistic regression benchmark: predicting a whole distribution, scored by CRPS and CRLS](https://www.ozercemkelahmet.com/model/bench-probabilistic-regression.md): How a tabular foundation model predicts full distributions, not single numbers, on ScoringBench, and how calibration fixed the CRLS gap. - [Metadata independence benchmark: predicting with no column names](https://www.ozercemkelahmet.com/model/bench-metadata-independence.md): How a tabular foundation model keeps its accuracy when column names are removed or scrambled, and how I measured that fairly against rivals. - [Blind sector identification: naming a dataset's industry from raw cells](https://www.ozercemkelahmet.com/model/bench-sector-identification.md): How a tabular foundation model names the industry of an unseen table from cell values alone, with no headers, labels or metadata, and how I tested it. - [Missing-value imputation benchmark: filling the gaps in real tables](https://www.ozercemkelahmet.com/model/bench-imputation.md): How a tabular foundation model fills missing numeric and categorical cells, and how a hunt for fifteen measurement errors produced a fair first place. - [Entity and schema matching benchmark: joining tables that share no keys](https://www.ozercemkelahmet.com/model/bench-entity-schema-matching.md): How a tabular foundation model matches records and columns across tables with no keys and no column names, and how I measured it against name-aware rivals. - [Missing-data robustness benchmark: predicting when the features have holes](https://www.ozercemkelahmet.com/model/bench-missing-data-robustness.md): How a tabular foundation model keeps its predictions when values are missing from the features, and why I checked the ruler before trusting any number. - [One checkpoint, then compression: packaging a whole benchmark result](https://www.ozercemkelahmet.com/model/bench-unified-checkpoint.md): How the core, every head and every adapter were merged into one checkpoint of about 2.18 billion parameters, then compressed and audited before release. - [Quickstart](https://www.ozercemkelahmet.com/model/quickstart.md): Get an API key, quick-run the model across sources that share no key, and publish the same configuration as an endpoint, in about five minutes. - [Introduction](https://www.ozercemkelahmet.com/model/introduction.md): What the model is, what one call returns, and the two ways to consume it. - [Authentication](https://www.ozercemkelahmet.com/model/authentication.md): - [Models](https://www.ozercemkelahmet.com/model/models.md): The foundation models you can run on and create endpoints on, how to choose a base, and how an upgrade moves an endpoint to a newer base. - [Data](https://www.ozercemkelahmet.com/model/data.md): - [Outputs](https://www.ozercemkelahmet.com/model/outputs.md): The one response every pass returns: sector, column profile, cross-table map, imputation, predictions, and on endpoints the held-out block. - [Endpoints](https://www.ozercemkelahmet.com/model/endpoints.md): Create an endpoint over pinned data, serve it with a serve-scoped key, refresh when the source changes, and upgrade when a new base ships. - [Held-out reports](https://www.ozercemkelahmet.com/model/held-out.md): How the held-out score is computed by in-context split evaluation, what every report contains, and exactly what the number claims. - [Agents and LLMs](https://www.ozercemkelahmet.com/model/integrations/agents.md): Give your agent the model as a tool, over MCP from any framework or over REST with one function-calling definition, and narrate the bundle with your own LLM. - [MCP server](https://www.ozercemkelahmet.com/model/mcp.md): Use the model from your AI. One remote server for Claude, ChatGPT, Cursor, VS Code, Gemini CLI, and any MCP client. Sign in or use an API key. Everything on the platform and the API, with the same output bundle. - [Workflows and data stack](https://www.ozercemkelahmet.com/model/integrations/data-stack.md): Call the model from any HTTP node, schedule syncs and refreshes from your orchestrator, and land bundles in the warehouse. - [Troubleshooting](https://www.ozercemkelahmet.com/model/troubleshooting.md): Every response the API returns when something needs your attention, what it is telling you, and the concrete next step. - [Glossary](https://www.ozercemkelahmet.com/model/glossary.md): The product vocabulary, one entry per term, identical across the API and the platform. ## Articles - [What is a Data Language Model?](https://www.ozercemkelahmet.com/articles/what-is-a-data-language-model.md): A foundation model that reads tables directly, with no shared keys, labels, or preprocessing. - [What is a tabular foundation model?](https://www.ozercemkelahmet.com/articles/what-is-a-tabular-foundation-model.md): A tabular foundation model is pre-trained on many tables and predicts, imputes, and detects anomalies on a new table without per-dataset training. - [Why build a large tabular foundation model?](https://www.ozercemkelahmet.com/articles/large-tabular-foundation-model.md): What scale adds to a tabular foundation model: one checkpoint for many tasks, more than 1.5 billion parameters, and the trade-offs of serving it. - [Real-time AI: from event streams to decisions](https://www.ozercemkelahmet.com/articles/real-time-ai-from-streams-to-decisions.md): What real-time AI is, how Apache Kafka and Apache Flink feed it, and how online and continual learning keep models current. - [Retrieval-augmented generation (RAG): from naive to agentic](https://www.ozercemkelahmet.com/articles/rag-from-naive-to-agentic.md): How retrieval-augmented generation works, why naive RAG fails, and how advanced, graph, and agentic RAG fix it, with evaluation. - [AI and machine learning for tabular data: a practical guide](https://www.ozercemkelahmet.com/articles/ai-and-machine-learning-for-tables.md): How AI, machine learning, and deep learning relate, which learning paradigms apply to tables, and when to choose trees or a foundation model. - [One pass, one bundle: what a table model returns](https://www.ozercemkelahmet.com/articles/one-pass-six-outputs.md): Sector, column profile, cross-table map, imputation, predictions, and a held-out block. - [Joining tables that share no key](https://www.ozercemkelahmet.com/articles/joining-tables-without-a-key.md): How a model can align columns and match rows across sources, and why the result should be a proposal. - [A score is only as good as its report](https://www.ozercemkelahmet.com/articles/a-score-is-only-as-good-as-its-report.md): What a held-out report should contain, and how to read one without being fooled by a headline number. - [Quick runs and endpoints: stateless and stateful](https://www.ozercemkelahmet.com/articles/quick-runs-and-endpoints.md): Two ways to use a table model: a one-shot call that stores nothing, and a persistent endpoint that holds state. - [Snapshots: freezing data so numbers stay meaningful](https://www.ozercemkelahmet.com/articles/snapshots-keep-numbers-meaningful.md): Why immutable data snapshots make reports, upgrades, and refreshes trustworthy. ## Projects - [Tabular foundation model](https://www.ozercemkelahmet.com/model.md): A model with more than 1.5 billion parameters that reads raw tables directly. - [Urban zoning intelligence](https://www.ozercemkelahmet.com/projects/urban-zoning-intelligence.md): A multi-city platform that makes zoning and property data queryable with LLMs and retrieval. ## Tools and methods - [Tools and methods](https://www.ozercemkelahmet.com/tools.md): long guides for the tools, methods, and models used. ## About - [About](https://www.ozercemkelahmet.com/about.md): work on large tabular foundation models, Data Language Models, and real-time AI.