Isotopes AI has emerged from stealth with a $20 million seed round and a clear thesis: business leaders shouldn’t have to wait on data engineers to answer fundamental questions. Its centerpiece, an AI agent named Aidnn, aims to turn plain-English requests into trustworthy, end-to-end analyses that span messy, distributed enterprise data.
The promise is bold but focused—close the chasm between the people who run data infrastructure and the people who need decisions. Instead of just chatting over dashboards, Aidnn is built to fetch, clean, reconcile, and synthesize data across systems, then present conclusions, caveats, and next steps.
An agent built for messy, real-world data
Where most “chat with your data” tools stop at SQL generation, Aidnn orchestrates multi‑step workflows. Ask for monthly recurring revenue by segment, and it can discover relevant sources in Snowflake or Databricks, pull subscriptions from finance platforms such as NetSuite or Stripe, join them with CRM data from Salesforce, clean inconsistent fields, prorate partial months, handle revenue recognition logic, and expose anomalies before calculating the final metric.
Crucially, the agent shows its work. Users can review the plan it drafted, the transformations applied, and the assumptions made. If something looks off—a spike from a one‑time adjustment, a suspicious mapping, a time window mismatch—Aidnn flags it and suggests remediation. The goal is trust by construction, not just a polished answer.
Founders with Hadoop-era scars
The company is led by Arun Murthy, a veteran of Yahoo’s original Hadoop team and a cofounder of Hortonworks, which later merged with Cloudera. After that chapter, he served as chief technology officer at Scale AI. Murthy is joined by longtime collaborators Prasanth Jayachandran and Gopal Vijayaraghavan—engineers who have watched enterprise data evolve from on‑prem clusters to cloud warehouses and still struggle to reach decision‑makers.
Their experience informs a simple diagnosis: the hard part isn’t the question—it’s the preparation. Years of investment in lakes, warehouses, and BI tooling haven’t eliminated the bottleneck of modeling, cleaning, and stitching together data for each new decision. IDC’s Global DataSphere has projected the world’s data to exceed 175 zettabytes by mid‑decade, while Anaconda’s State of Data Science reports practitioners still spend roughly 40% of their time on data prep. That gap is Aidnn’s target.
How Aidnn works under the hood
Behind the scenes, Aidnn blends large language models with tool use. It introspects catalogs and schemas, reasons over metadata, and drafts a stepwise plan: discover sources, validate lineage, generate SQL or Python for transformations, apply business rules, and assemble outputs as tables, charts, or narratives. It keeps long‑lived context so a planning conversation doesn’t reset every time a user pivots from top‑line to subsegment analysis.
The system emphasizes reproducibility. Plans can be inspected, versioned, and rerun, with derived datasets cached to cut latency on follow‑ups. It can work alongside semantic layers and dbt models where they exist, or construct temporary logic when they don’t. The company says the agent prioritizes data quality checks—type conformity, null patterns, deduplication—and ties insights to lineage so teams can trace where numbers came from.
Privacy, deployment, and governance
Enterprises are wary of sending proprietary data to external model providers. Isotopes says Aidnn can be deployed within a customer’s cloud environment and does not share customer data back to foundation model makers. Access controls, PII masking, and policy enforcement are designed to mirror existing governance, with full audit trails that record who asked what, which data was read, and what transformations were executed.
For regulated industries, that control is as important as accuracy. Gartner’s recent analytics leadership research continues to list data quality, lineage, and trust as top barriers to scaling AI. Aidnn’s “show your work” posture is meant to meet compliance teams where they live.
Crowded field, different bet
Isotopes enters a busy arena. Salesforce has been weaving agents into Tableau; Microsoft is infusing Copilot across Fabric; Snowflake and Databricks have launched their own assistants; ThoughtSpot and others have long chased natural‑language BI; startups like WisdomAI are pursuing agentic analytics as well. The differentiation Isotopes is pitching is not a prettier chat window, but an agent that takes responsibility for data preparation and modeling in context—what amounts to on‑demand, task‑specific ETL with a paper trail.
Backing for the $20 million seed came from NTTVC, led by investor Vab Goel, alongside other supporters with deep enterprise pedigrees. That capital will be aimed at connectors, governance features, and the unglamorous reliability work that determines whether an agent becomes a daily tool or a demo.
Why this matters for enterprises
Consider a recurring headache like revenue reporting. Most companies have billing in Stripe or Zuora, contracts in NetSuite, customer hierarchies in Salesforce, and product data in a warehouse. Finance and sales operations spend days reconciling edge cases—partial upgrades, mid‑cycle cancellations, currency conversions—before a single MRR chart is safe to share. An agent that can reliably execute that chain, expose assumptions, and keep context across weeks of planning cycles would compress time to insight and reduce manual risk.
That is the bet behind Aidnn: shift analytics from ad hoc heroics to repeatable, explainable workflows that anyone can initiate. If Isotopes can make that dependable at scale—without leaking data or inventing numbers—it will have answered one of big data’s oldest questions: how to put the right data, in the right shape, into the hands of the people who need it most.