--- title: Rishabh Patil emoji: 🔎 colorFrom: indigo colorTo: blue sdk: static pinned: true short_description: AI engineer and researcher. Models, data, papers. tags: - portfolio - claim-verification - text-to-sql ---

AI engineer and researcher in London. Co-founder and CEO of [Valuren](https://www.linkedin.com/company/valuren/), which gives fashion and luxury brands digital product passports. I work on verification for AI: checking claims against evidence, evaluating LLM systems, and planning safely in learned latent spaces. The page at the top of this Space pulls live download numbers from the Hub API. ## VeriSci: scientific claim verification

Paste a scientific claim and get SUPPORTS, REFUTES or NOT ENOUGH INFO, with the sentences that justify the label. VeriSci pipeline: claim, retrieve with BM25 and a fine-tuned e5 retriever, select evidence sentences, verify with a calibrated DeBERTa model, return a label with evidence. Retriever Recall@5 0.867. Verifier accuracy 0.905, macro-F1 0.874, calibration error 0.024. End to end on SciFact validation: 0.691 accuracy, 0.653 macro-F1.

VeriSci pipeline: claim, retrieve with BM25 and a fine-tuned e5 retriever, select evidence sentences, verify with a calibrated DeBERTa model, return a label with evidence. Retriever Recall@5 0.867. Verifier accuracy 0.905, macro-F1 0.874, calibration error 0.024. End to end on SciFact validation: 0.691 accuracy, 0.653 macro-F1.

| Piece | Repo | | :-- | :-- | | Live demo | [verisci-claim-space](https://huggingface.co/spaces/rishhh/verisci-claim-space) (may take a minute to wake if idle) | | Release verifier | [verisci-claim-verifier-retrieval-adapted-seed123](https://huggingface.co/rishhh/verisci-claim-verifier-retrieval-adapted-seed123) | | Retriever | [verisci-scifact-e5-retriever](https://huggingface.co/rishhh/verisci-scifact-e5-retriever) | | Ablations | Seeded variants of evidence selection (joint, evidence-gated, hard-negative, contradiction-adapted), each with its own model card | ## SchemaSage-SQL: safe text-to-SQL

QLoRA adapters on Qwen3-4B that write read-only SQL grounded in the schema you give them, with a safety layer that refuses destructive requests. SchemaSage-SQL release baseline on 64 cleaned held-out examples: 100% SQL parse validity, 98.3% schema adherence, 0% unsafe queries, 100% correct refusals. QLoRA on Qwen3-4B-Instruct-2507, trained on 111,444 cleaned examples, 10,862 of them refusals.

SchemaSage-SQL release baseline on 64 cleaned held-out examples: 100% SQL parse validity, 98.3% schema adherence, 0% unsafe queries, 100% correct refusals. QLoRA on Qwen3-4B-Instruct-2507, trained on 111,444 cleaned examples, 10,862 of them refusals.

[Preprint](https://doi.org/10.5281/zenodo.20497841) | | [clap-family](https://github.com/MrRobotop/clap-family) | Conservative Lapse-Action Planning for safe latent trajectory optimisation, 72 tests | PyPI version

[Preprint](https://doi.org/10.5281/zenodo.20467271) | | [micro-world-model](https://github.com/MrRobotop/micro-world-model) | Hierarchical JEPA world model that plans in latent space without reconstruction or reward | [Preprint](https://doi.org/10.5281/zenodo.20480620) | | [ariadne-eval](https://github.com/MrRobotop/ariadne-eval) | Tracing and scoring for multi-step LLM agents, 285 tests | Tests, live status

| | [evalforge](https://github.com/MrRobotop/evalforge) | LLM evaluation with bootstrapped confidence intervals and a CI regression gate, 538 tests | Tests, live status

| ## Background MSc Artificial Intelligence, University of St Andrews (dissertation on dependent types for machine learning, supervised by Dr Edwin Brady). BSc Artificial Intelligence, VU Amsterdam. [GitHub](https://github.com/MrRobotop) · [LinkedIn](https://www.linkedin.com/in/rishabh-ashok-patil/) · [ORCID](https://orcid.org/0009-0007-0868-9673) · [Website](https://www.rishabhpatil.com) · [Email](mailto:rishabh.a.patil@outlook.com)