
Share
Most AI data programs stall not on infrastructure but on tribal knowledge trapped in spreadsheets and veteran employees' heads. DataSwitch's new Mathi engine bets that automating those judgment calls, not just the pipes, is the real unlock.
DataSwitch has introduced Mathi, an AI-powered judgment engine designed to solve a problem that has quietly throttled enterprise AI adoption for years: the gap between what experts know and what data pipelines can execute. The Lewes, Delaware and Chennai, India-based company positions Mathi as the decision-making core of its Expert-to-Pipeline (E2P) platform, which converts domain expertise into production-ready data pipelines proposed in minutes, approved by experts, and live within days.
The thesis here is narrow but credible. Legacy data tooling moves data from point A to point B. It does not decide which fields correspond to which, or which business rules govern a transaction. Those decisions have historically required months of spreadsheet work and workshop sessions between engineers and subject-matter experts. Much of that institutional knowledge never gets documented. It lives in people's heads or buried in legacy code that nobody wants to touch.
A global insurance group is already running Mathi in production, using it to integrate acquired business units and legacy policy data. Trained on the insurer's ontology covering claims, policy types, underwriting logic, and regulatory requirements, Mathi recommends mappings from regional source schemas into standardized target models. Underwriting and actuarial experts review and approve each recommendation before DataSwitch generates pipelines and validation tests. The stated goal is to compress source onboarding from months into days, freeing human experts to focus on exceptions rather than routine mapping decisions.
Mathi's mechanics are built around a feedback loop rather than a one-shot automation pass. The system predicts data mappings, derives business rules from existing code, applies domain-specific ontologies, and documents legacy logic that would otherwise remain opaque. Every recommendation carries a confidence score and supporting evidence, which gives human reviewers something concrete to evaluate rather than a black-box output.
Experts approve or reject decisions in plain language. Each approval feeds back into the system, improving subsequent recommendations. Once a decision is approved, it automatically generates pipelines, validation tests, and lineage documentation. That lineage piece matters for regulated industries like insurance and healthcare, where auditability is not optional.
"Every data program starts with decisions that live in experts' heads, and those decisions are still made by hand," said Karthikeyan Viswanathan, CEO and President of DataSwitch. "That is the real bottleneck to AI readiness. Legacy tools move data; DataSwitch automates data decisions. With Expert-to-Pipeline, experts stay in control, knowledge stays inside the enterprise, and work that took months goes live in days."
One detail worth flagging for risk-conscious buyers: Mathi runs on Medha, DataSwitch's proprietary small language model, and operates air-gapped on standard CPUs without external model calls. No data leaves the enterprise perimeter, and no third-party API dependency exists. For insurance, healthcare, and life sciences customers handling sensitive policyholder and patient data, that architecture choice removes a category of vendor risk that comes with cloud-hosted large language models. It also sidesteps GPU procurement constraints that have slowed other AI deployments across the industry.

Within E2P, Mathi's approved decisions direct AI agents that build native pipelines for platforms including Databricks, Snowflake, and Microsoft Fabric. The resulting data is meant to feed downstream analytics and AI applications such as Databricks Genie and Snowflake Cortex. The whole stack sits on top of DataSwitch's existing SwitchIE engine, which the company says already supports more than 30 technologies and works with over 45 strategic system integrator partners.
DataSwitch cites customer results of up to 70% lower modernization costs and 10x faster migrations, though these figures come from the company's own marketing materials rather than independent audits. The company holds ISO/IEC 27001 certification, a baseline credential for enterprise data security buyers evaluating vendors with access to sensitive systems.
The timing lines up with a broader industry conversation about why enterprise AI projects stall. A companion piece published on the same outlet the prior day explored knowledge graphs and large language models grounding enterprise answers in real dependencies, a parallel attempt to solve the same underlying issue: AI systems need structured, trustworthy context about how a business actually operates, not just raw data volume. Another recent piece on unstructured data noted that enterprises spent two decades perfecting structured data management while AI now demands answers from the messier 80 to 90% of data that was never organized in the first place.
DataSwitch's bet is that the bottleneck is not compute, not storage, and not even data volume. It is the slow, manual process of encoding tacit expertise into rules a machine can execute reliably. If that diagnosis is correct, demand for judgment-automation tools like Mathi should track closely with how many large enterprises are stuck in extended data modernization cycles, which based on anecdotal industry commentary is a sizable population.
The insurance deployment is a single customer reference, and DataSwitch has not disclosed contract value, deployment timeline specifics beyond the "days not months" framing, or whether the 70% cost reduction and 10x migration speed figures apply broadly or only to select use cases. Buyers and investors should press for independently verified case studies before treating these numbers as representative.
Competitive context also matters. SnapLogic's recent MCP Discovery Hub launch and other agent-building and governance tools suggest the market for AI-ready data infrastructure is getting crowded, with multiple vendors converging on similar value propositions from different architectural angles. DataSwitch's differentiation rests on the air-gapped small language model and the expert-approval workflow, but whether that holds up against larger, better-capitalized competitors with broader platform reach remains an open question. Watch for additional customer disclosures, particularly outside insurance, as the clearest signal of whether Mathi's approach generalizes.
Tags
Original Sources
DataSwitch Introduces Mathi to Turn Enterprise Expertise into AI-Ready Data Pipelines - BigDATAwire
↗ https://www.hpcwire.com/bigdatawire/this-just-in/dataswitch-introduces-mathi-to-turn-enterprise-expertise-into-ai-ready-data-pipelines
About the author
Marcus began tracking AI's market implications in 2016, noticing AI-related patent filings accelerating ahead of earnings upgrades before most of the sell-side had caught on. A former fixed-income quantitative analyst, he spent two decades building models that priced risk across emerging markets before pivoting to cover the economic impact of AI full-time. His writing translates opaque technical developments into clear risk/reward terms — and he's rarely diplomatic about the gap between AI valuations and underlying fundamentals. He believes most market participants still underestimate AI's long-run deflationary effect on knowledge work.
More from The Analyst →This Week's Edition
18 November 2026
29 articles
Related Articles

Elsevier Gives ClinicalKey AI Full EHR Access, Betting on Patient Context Over Chatbots
Products & Applications · 6 min

BMO Claims Quantum Utility Milestone in AI Feature Selection, Using 100x Less Data
Products & Applications · 5 min

National Academies Report Calls for Faster, Tougher Retraction Standards
Models & Research · 5 min
Related Articles

Elsevier Gives ClinicalKey AI Full EHR Access, Betting on Patient Context Over Chatbots
Products & Applications · 6 min

BMO Claims Quantum Utility Milestone in AI Feature Selection, Using 100x Less Data
Products & Applications · 5 min

National Academies Report Calls for Faster, Tougher Retraction Standards
Models & Research · 5 min
More Stories
© 2026 Cedar & Bloom. All rights reserved.