Ayush Rana
AI Engineer · Founder & CTO
I'm Ayush Rana. I build AI systems for places where "the model said so" is not an acceptable answer — banks, NBFCs, insurers, anyone whose numbers have to survive an audit. I founded Warelytics, where I'm building Daufx: an AI data analyst that answers questions in plain English, never lets a raw row leave the customer's infrastructure, and seals every analysis with a replayable cryptographic attestation. Before that I spent three years at KPIT building web and cloud applications in Angular, Java and AWS — enterprise software with real users behind it, which is where I learned that shipping a feature is the easy part and standing behind it afterwards is the actual job.
I'm based in Bhopal, Madhya Pradesh, India, and I speak English and Hindi.
What I'm doing now
I'm the Founder & Chief Technology Officer of Warelytics AI Solutions Pvt. Ltd., which I founded in September 2025 in Bhopal, Madhya Pradesh, India. A governed AI data analyst for regulated financial institutions — private by architecture, auditable by design, live on the customer's own data.
Our product is Daufx, currently in active development. The short version: analysts get to ask questions in plain English, the firm's raw data never leaves its own infrastructure, and every answer arrives with a replayable record of how it was produced.
Where I've worked
Founder & Chief Technology Officer — Warelytics AI Solutions Pvt. Ltd.
September 2025 to Present · Bhopal, Madhya Pradesh, India
I founded Warelytics to solve one problem properly: regulated firms want AI analytics and cannot legally paste their data into a public model. Daufx is the answer — it runs inside the customer's boundary, shows the planning model only masked schema and aggregates, and seals every analysis with an attestation the compliance team can replay.
- Designed the privacy bridge: sensitive data is cleaned, profiled and analysed by local models on customer infrastructure; the cloud planning model never receives a raw row.
- Built the attestation layer — every analysis emits a hash-chained record of plan, code, tools and outputs, replayable on demand.
- Shipped connectors for PostgreSQL, MySQL, S3-compatible object storage, Azure, Kafka and flat files, with change-data-capture so answers reflect live state rather than stale exports.
- Deployment across managed cloud, customer VPC in AWS/Azure India regions, and fully air-gapped on-premise, with SSO/SAML, column-level access control, BYOK encryption and SIEM audit export.
- Targeted the compliance surface deliberately: RBI, SEBI and IRDAI reporting, and the DPDP Rules notified in November 2025 with full compliance due May 2027.
Software Engineer — KPIT
October 2023 to January 2026 · Pune, Maharashtra, India · Hybrid
Full-stack and cloud application engineering — Angular on the front, Java and AWS services behind it. Two and a bit years of building software other people depended on, inside a large engineering organisation with the review cycles and release discipline that implies.
- Built and maintained Angular front-ends for internal and customer-facing applications.
- Worked across AWS services for hosting, storage and integration, and on the Java services behind the UI.
- Delivered inside an enterprise process: code review, structured releases, and long-lived codebases where the person maintaining your work in a year might not be you.
Worked with: Angular, AWS, Java, TypeScript.
Intern — KPIT
January 2023 to September 2023 · Pune, Maharashtra, India · On-site
Nine months learning to write software that other people have to read. Java and Angular, on real codebases rather than exercises — which is a different experience from either, and the reason the graduate role came next.
- Java and Angular development on production codebases.
- First exposure to working inside an established engineering team and its conventions.
Worked with: Java, Angular.
What I'm interested in
- AI that can be audited
- The interesting engineering problem in AI right now is not capability, it is accountability. Anyone can get a model to produce a number. Producing a number you can hand to a regulator, with a record of exactly how it was derived, is a different job.
- Systems with hard constraints
- I like requirements that cannot be negotiated with. Data residency, air-gapped networks, audit trails, a security review you either pass or fail — constraints make a lot of the design decisions for you, and I would rather have them than a blank page.
- Local and small models
- A lot of what people reach for a frontier API to do can be done by a small model running on the customer's own hardware, for a fraction of the cost and none of the data-exit problem. Working out which is which is most of the value.
- Data that arrives broken
- Real enterprise data is inconsistent, badly typed, and contradicts itself across systems. Profiling and grading it honestly before anyone builds on top is unglamorous and it is where most AI projects quietly fail.
- Building the company as well as the product
- Deciding what not to build, what an evaluation has to prove before anyone signs, and how to survive a bank's security review are engineering problems too. They just have different failure modes.
- Explaining hard things simply
- If a compliance officer with no ML background cannot follow how a number was produced, the architecture is not finished. That constraint has improved my systems more than any framework.
Things I've built
Daufx — An AI data analyst a compliance team will actually approve (Product, 2026)
Status: In active development.
The thing I've spent the most time on. Analysts at regulated firms want to ask questions in plain English; their regulator wants every reported number to be reconstructable. Daufx does both. Sensitive data is profiled and analysed by local models inside the customer's own infrastructure, the cloud planning model only ever sees masked schema and aggregates, execution happens in a sandbox, and the whole run is sealed as a hash-chained attestation that can be replayed on demand. It connects straight to live databases with change-data-capture, so answers reflect reality rather than last month's export.
Built with: Local + frontier LLMs, Column-level PII masking, Cryptographic attestation, Change-data-capture, PostgreSQL, Kafka, Sandboxed execution, On-prem / VPC / air-gapped.
The attestation layer — Every answer carries a receipt (Platform, 2026)
Status: Shipping inside Daufx.
The part I find most interesting. An analysis is not just its output — it is the plan, the generated code, the tools invoked, and the data version it ran against. Daufx hash-chains all of it into a tamper-evident record, so an examiner asking "how did you arrive at this figure" gets a replayable answer instead of somebody reconstructing it from memory and a spreadsheet. Audit preparation stops being archaeology.
Built with: Hash chaining, Replayable execution, Audit trails, Tamper evidence.
The privacy bridge — Use a frontier model without letting it see the data (Platform, 2026)
Status: Shipping inside Daufx.
Regulated firms are stuck: the good models are hosted, and their data is not allowed to leave. The bridge splits the problem — a local model does anything that requires touching real rows, and the hosted planning model receives only masked column names and aggregate shapes. It plans; the local side executes. The customer gets frontier-quality reasoning without a single raw row crossing their boundary, which is the difference between a security review passing and failing.
Built with: Local inference, Schema masking, Aggregate-only planning, Data residency.
Sentinel — Analytics that doesn't wait to be asked (Product, 2026)
Status: In design.
Where this goes next. Instead of answering questions, Sentinel watches connected data, notices when something warrants investigation, and then generates and answers its own questions against live retrieval — pulling in outside context where it genuinely changes the analysis. Every question it asks itself is logged and attested exactly like a human-asked one, which is the only way a proactive system stays auditable.
Built with: Agentic monitoring, Live retrieval, Attested runs, Anomaly detection.
Three years of enterprise software at KPIT — Angular, Java and AWS, on codebases that outlive you (Engineering, 2023–2026)
Status: 3 years, Pune.
Where I learned the craft. Full-stack and cloud work — Angular front-ends, Java services, AWS underneath — inside a large engineering organisation, which means code review, structured releases, and codebases that will still be running long after you have moved on. Nine months of it as an intern first. It is the least glamorous entry on this page and the reason the rest of it works: writing software that a stranger has to maintain teaches you more about design than any amount of building things alone.
Built with: Angular, Java, AWS, TypeScript.
What I can do
AI engineering
- Agentic AI development
- Retrieval-augmented generation
- Local & small-model deployment
- Prompt and context engineering
- Evaluation & guardrails
- Cost/latency optimisation
Architecture
- Enterprise architecture
- Privacy-preserving system design
- Data residency & tenancy isolation
- Auditability & attestation
- API design
Data
- PostgreSQL
- MySQL
- Kafka
- Change-data-capture
- Parquet & object storage
- Data profiling & quality grading
Platform & deployment
- AWS
- Azure
- Docker
- On-premise & air-gapped delivery
- SSO/SAML & RBAC
- BYOK encryption
Application engineering
- Angular
- TypeScript
- Java
- Full-stack development
- Enterprise release process
Regulated delivery
- RBI / SEBI / IRDAI reporting context
- DPDP Act readiness
- Security review & vendor onboarding
Questions people ask
- Who is Ayush Rana?
- Ayush Rana is an AI engineer based in Bhopal, Madhya Pradesh, India, and the founder and Chief Technology Officer of Warelytics AI Solutions Pvt. Ltd. He builds governed AI systems for regulated industries — analytics that runs inside a customer's own infrastructure and produces auditable, reconstructable answers. Before founding Warelytics in September 2025 he spent three years at KPIT in Pune as a software engineer, building web and cloud applications in Angular, Java and AWS.
- What does Ayush Rana build?
- Governed AI for organisations that cannot send their data to a public model. His main work is Daufx, an AI data analyst for banks, NBFCs and insurers: users ask questions in plain English, sensitive data is processed by local models inside the customer's own boundary, the hosted planning model only ever sees masked schema and aggregates, and every analysis is sealed as a replayable cryptographic attestation. He also designed the privacy bridge and attestation layer underneath it.
- What is Warelytics?
- Warelytics AI Solutions Pvt. Ltd. is an AI company founded by Ayush Rana in September 2025, based in Bhopal, India. Its product, Daufx, is a governed AI data analyst for regulated financial institutions, currently in active development — designed to deploy in managed cloud, a customer's own VPC in AWS or Azure India regions, or fully air-gapped on-premise, with column-level PII masking and tamper-evident audit trails.
- Why does governed or auditable AI matter?
- Because in regulated industries an answer without provenance is unusable. Indian regulators including the RBI, SEBI and IRDAI expect reported figures to be reconstructable, and the DPDP Rules notified in November 2025 carry penalties of up to ₹250 crore for security-safeguard failures with full compliance due by May 2027. Pasting data into a public AI tool is untracked third-party processing the firm remains liable for. Daufx exists so analysts get AI assistance without the firm taking on that exposure.
- What is Ayush Rana's background before AI?
- Enterprise software engineering. He spent three years at KPIT in Pune, Maharashtra — nine months as an intern from January 2023, then as a software engineer from October 2023 to January 2026 — building web and cloud applications with Angular, Java and AWS. He credits that period with learning to write software other engineers have to maintain, and with the emphasis he now places on verification and reproducibility in AI systems.
- Is Ayush Rana available for consulting or contract work?
- He is contactable through his site and through LinkedIn. His focus is Warelytics and Daufx, and the work he is most useful for is adjacent to it: privacy-preserving AI architecture, retrieval systems over messy enterprise data, and getting AI deployments through security and compliance review in regulated environments.
- How can I contact Ayush Rana?
- By email at ayush.tech8187@gmail.com, or through LinkedIn. Both are linked from his site, along with the Warelytics company site at warelytics.ai.
Get in touch
Email: ayush.tech8187@gmail.com
Company: warelytics.ai
Website: ayushrana.dev
LinkedIn: linkedin.com/in/ayush-rana-87b158200