Arya.ai

Where Arya.ai sits

Arya.ai is an enterprise AI platform built specifically for banking, insurance and lending — not a general-purpose AI product with a finance skin. Founded in 2013 in Mumbai by IIT Bombay researchers Vinay Kumar Sankarapu and Deekshith Marla, it was one of the first deep-learning startups to push models into regulated financial institutions, and since April 2024 it has been majority-owned (67%, roughly $16.5 million all-cash) by Aurionpro Solutions, a listed enterprise software firm in banking and financial services. That ownership matters for buyers: the startup independence question is answered, the balance sheet is public, and the product roadmap now sits inside a banking-software vendor's portfolio rather than a venture-funded sprint. The vendor states more than 100 banks and insurers use the platform, concentrated in India and Southeast Asia, with named customers including ICICI Lombard, Axis Bank, TATA AIG General Insurance and MSIG Life.

The pitch is verticalization. Where horizontal AI platforms hand you a model and wish you luck with compliance, Arya ships pre-trained models for KYC extraction, document fraud detection, bank statement analysis, invoice parsing, credit risk decisioning and claims processing — tasks with names a risk officer recognizes — and wraps them in deployment options (cloud, on-premise, hybrid) and governance tooling built for institutions where an unexplainable model is a regulatory incident. This is a sales-led enterprise product, and I reviewed it as one: what the platform actually contains, what you can evaluate without a contract, and where the opacity begins.

What it actually does

The product family has four named pieces. Apex is the AI API library: the vendor counts more than 100 pre-trained APIs covering customer identity verification, document fraud detection, invoice and KYC extraction, bank statement analysis, resume parsing, QR reading and adjacent tasks, sold on pay-as-you-go credits where one credit equals one successful API call. Nexus is an API gateway for traffic management, governance and compliance controls on top of those APIs. Prism is a secure generative-AI workspace for teams, currently marked "coming soon." And Infinity OS is the newest layer — an agentic workflow canvas where enterprises wire together pre-trained applications, frontier models (the vendor names ChatGPT, Claude and Llama as model-agnostic options), 200+ OAuth integrations, human-in-the-loop checkpoints and full execution audit trails, deployable to private cloud for regulated work.

The self-serve story is Apex. Register with a verified email, subscribe to an API module, and you get a token and documentation. Every subscription includes a free trial of 50 API calls, and registering with a corporate email adds 500 free calls across five modules (KYC, Face to BMI, Emotion Analysis, Face Verification, Salaryslip Extraction). Authentication is a per-user secret token passed in the request header. Beyond that, pricing is pay-as-you-go at the API layer and quoted per institution for platform deployments — there is no public price list for Nexus, Infinity OS or enterprise agreements.

Compliance positioning is central and concrete: the vendor states no data is stored from API processing, systems follow ISO 27001 guidelines (the Apex page cites ISO/IEC 27001:2022), GDPR compliance is claimed, and there is a public status page at status.arya.ai. Vendor performance claims — "95%+ average AI accuracy," "300M+ annual API calls," "99% API success rate" — are marketing figures I could not independently verify and would want contractual definitions around before signing.

My testing notes

My first observation is about the onboarding flow, which is unusually testable for an enterprise product. The free trial credits mean a developer can subscribe to a KYC extraction API and make real calls the same afternoon — no procurement call, no demo gate. The documentation is organized per subscribed module with endpoint, token and request/response formats, and the token-in-header scheme is simple to implement. Where it shows its age is in the developer experience around the edges: the credit model (buy credit blocks per API, top up in the Subscriptions tab) is manual compared to modern usage-based billing, and there is no evidence of the sandbox fixtures, SDKs or typed clients that self-serve API platforms now treat as table stakes.

Second, the vertical depth is the real differentiator and it is visible in the module list. Generic "document extraction" APIs are everywhere; a bank statement analyzer that understands varied Indian bank formats, a salary-slip extractor, and document tampering detection are not. The customer testimonial from ICICI Lombard's operations chief — claims approval times cut from 60 minutes to under a minute — is a vendor-published quote, but it is the kind of specific, attributable claim that matches the problem set the platform is built for, and the customer list (Axis Bank, TATA AIG, HDFC Bank divisions) is corroborated across the Aurionpro acquisition coverage and the vendor's own pages.

A fourth point on the human side: the testimonial wall on the vendor pages is unusually specific for enterprise marketing — named operations and technology chiefs at ICICI Lombard, TATA AIG and MSIG Life describing concrete workflow changes, plus a stated month-long go-live for a high-volume API integration at one insurer. Quotes like these are curated by the vendor, but specificity of name, role and outcome is a better trust signal than anonymous logos, and it is the kind of claim you can test in a reference call.

Third, the roadmap is moving fast and partially unfinished. Infinity OS is the flagship story on the homepage — agentic workflows, human checkpoints, audit trails, private cloud — and it reads like a strong answer to the "explainable AI" demand from regulators. But Prism is explicitly "coming soon," Infinity OS's maturity signals are thin outside marketing pages, and the products under the Aurionpro umbrella (AryaXAI observability, Libra fine-tuning from the acquisition era) have been folded or renamed in ways that are hard to follow from the outside. In enterprise buying, roadmap coherence is a reliability signal, and I would want a written product commitment before betting a workflow on the newest layer.

Limits

There is no public pricing beyond Apex credits — every platform deployment is a conversation with sales, which makes cost comparison against alternatives impossible without an RFP. The self-serve surface is a thin developer portal: no visible SDK ecosystem, no community forum, no public issue tracker. The accuracy and volume claims (95%+ accuracy, 99% API success, 300M+ calls) are vendor assertions without published benchmarks or methodology. Coverage is India/Southeast Asia-centric; document formats and compliance packs are tuned to those markets first. Prism does not exist yet as a shipped product. And the per-API credit model fragments billing: a workflow spanning six APIs means six subscriptions and six credit pools to manage rather than one metered bill.

Verdict

Arya.ai is a serious, well-anchored vertical AI platform for financial institutions — production-ready models for named banking and insurance tasks, real deployment flexibility, a compliance posture built for auditors, and the backing of a listed parent since 2024. It is also a sales-led enterprise product with opaque platform pricing, vendor-sourced performance claims, and a roadmap that is visibly mid-refactor around Infinity OS. Rated 3.7 on the rubric: the core functionality is deep and genuinely vertical for BFSI workflows, though parts of the platform stack are still maturing (40% core functionality), value is hard to judge without quotes but the free Apex trials de-risk evaluation (20% value), the low-code API approach is approachable for developers while enterprise onboarding remains manual (20% usability), named enterprise customers and institutional ownership are strong ecosystem signals (10%), and ISO 27001 claims, a public status page and a profitable-acquisition history are good reliability evidence set against unverifiable accuracy figures and a partially shipped roadmap (10%). If you are a bank, insurer or lender in Asia evaluating pre-trained financial AI, this belongs on the shortlist alongside Hyperscience and DataRobot-class vendors. If you are a startup wanting self-serve document AI, the friction will outweigh the fit.

Sources

• https://arya.ai/
• https://arya.ai/apex
• https://api.arya.ai/faqs
• https://arya.ai/ai-native-os
• https://arya.ai/industry/banking
• https://arya.ai/industry/insurance
• https://arya.ai/about-us
• https://www.aurionpro.com/news/aurionpro-solutions-acquires-arya-ai-to-power-next-generation-enterprise-ai-platforms-for-financial-institutions-globally/
• https://economictimes.indiatimes.com/tech/startups/aurionpro-solutions-to-acquire-fintech-startup-arya-ai-for-165-million/articleshow/109416676.cms
• https://www.thehindu.com/news/cities/mumbai/business/aryaai-raises-750000-in-preseries-a-funding/article8020454.ece
• https://challengingvoice.com/tools/arya-ai
• https://forwardfuture.com/tools/details/apex-by-arya.ai