Isotopes AI has exited stealth mode with a $20 million seed round and a specific thesis: business leaders should not have to wait on data engineers to answer most questions. Its centerpiece, an AI agent called Aidnn, seeks to transform simple-English requests into reliable, end-to-end analytics across the spectrum of messy, distributed enterprise data.
The promise is brash but specific: bridge the gap between who runs data infrastructure and who needs decision-makers. Rather than simply chat across dashboards, Aidnn is designed to access, clean, reconcile, and integrate data from systems, then present findings, caveats and next steps.
An agent designed for messy, real-world data
Where most “chat with your data” tools leave you hanging with SQL generation, Aidnn runs multi‑step workflows. Request monthly recurring revenue by segment, and it can hunt down sources in Snowflake or Databricks, collect subscriptions from financial tools like NetSuite or Stripe, merge them with CRM data in Salesforce, cleanse the dirty fields, prorate partial months, apply revenue recognition logic and bubble up abnormalities before ripping the final metric.
Crucially, the agent shows its work. The users can see the plan generated by it, the transformations it carries on, the hypothesis it works under. If something appears funky — a spike caused by a one‑time adjustment, a shady looking mapping, a time window mis-match — Aidnn alerts and recommends corrective action to fix it. What we want is trust by construction, not only smooth talking.
Founders with Hadoop-era scars
The firm is run by Arun Murthy, who worked on Yahoo’s original Hadoop team and is a co-founder of Hortonworks, which was absorbed by Cloudera. He was C.T.O at Scale AI after that chapter. Murthy is working on the startup with two of his long-time engineering colleagues, Prasanth Jayachandran and Gopal Vijayaraghavan, who wanted to understand why enterprise data has evolved over the years from on‑prem clusters to cloud warehouses, but it was still so hard for constituents to get access to meaningful data.
Their experience leads to a simple diagnosis: The hard part isn’t the question. It’s the prep. Years of investment in lakes, warehouses and BI tooling hasn’t broken the bottleneck of modeling, cleaning, and joining data together for each decision. IDC’s Global DataSphere estimates that global data will reach over 175 zettabytes by mid‑decade, and Anaconda’s State of Data Science released in October shows that practitioners are still spending around 40% of their time on data prep. That gap is Aidnn’s target.
How Aidnn works behind the scenes
In the wings, Aidnn mixes large language models with tools. It introspects catalogs and schemas, reasons about metadata, and sketches a stepwise plan: find sources, validate lineage, handwrite SQL or Python for transformations, apply business rules, and assemble outputs as tables, charts, narratives. It retains long‑lived context so a planning conversation isn’t reset every time a user swivels from top‑line to subsegment analysis.
The system emphasizes reproducibility. They can be examined, versioned and rerun, with cached output of derived data sets to decrease latency on follow‑up investigations. It can co-exist with semantic layers and dbt models where they already exist, or build temporary custom logic where they don’t. The agent, the company says, also focuses on quality checks — type conformity, null patters, deduplication — and ties insights to lineage so it is possible to trace where numbers come from for teams.
Privacy, deployment, and governance
Businesses are reticent about sharing proprietary data with a third-party model provider. Isotopes weaves Aidnn into a cloud computing model for use within a customer’s cloud environment, and doesn’t feed customer data back to foundation model makers. Access controls, PII masking, policy enforcement is meant to replicate the current governance with full audit trails also available of who asked what, whose data was read, and what transformations ran.
For industries that are subject to regulation, that control can be as critical as accuracy. Gartner’s latest analytics leadership research classifiers data quality, lineage, and trust as top anathemas to scaling AI. What Aidnn’s “show your work” posture is really for is to meet the cheap amateur compliance guys where they are.
Crowded field, different bet
Isotopes enters a busy arena. Salesforce has also been infusing agents into Tableau; Microsoft has been incorporating Copilot throughout Fabric; Snowflake and Databricks have introduced their own assistants; ThoughtSpot and others have long been playing with natural‑language BI; startups like WisdomAI are seeking out agentic analytics, too. It’s not a better chat window that Isotopes is selling, it is an Agent that owns the data prep & model in context—the closest thing to on‑demand task specific ETL with a paper trail.
The $20 million seed was led by investor Vab Goel’s NTTVC, with participation from other enterprise folks with skin in the game. Share That capital will be pointed at connectors, governance features, and the unsexy reliability work that decides whether an agent is a daily tool or a demo.
Why this is important for businesses
Take a repeated pain point, similar to revenue reporting. In most companies, billing is in Stripe or Zuora, contracts in NetSuite, customer hierarchies in Salesforce, product data in a warehouse. Finance and sales ops spend days reconciling edge cases — partial upgrades, mid‑cycle cancellations, currency conversions — before a single MRR chart is trusted to see the light of day. An agent that can choreograph that chain reliably, to show assumptions, and to maintain context across weeks of planning cycles, would compress time to insight and manual risk.
That’s the bet behind Aidnn: moving from ad hoc heroics in data analytics to repeatable, explainable workflows that anyone can take on. If Isotopes can make that reliable at scale — without the data leaking or the numbers being made up — it will have solved one of the oldest problems in big data: how to put the right data, in the right shape, into the right hands.
