Short answers written to be quoted. The first group is about Octelion itself. The topic pages below go deeper on the work: lakehouses, Power BI, AI agents, workflows and monitoring.
Each page is a set of direct answers on one line of work, with the terms explained as they come up.
Medallion layers, data quality, governance, cost, migration from a warehouse, and what a handover contains.
12 questionsWhy reports are slow, what a star schema changes, incremental refresh, aggregations, Fabric, and the semantic model.
11 questionsWhat an agent is, what keeps it safe, evals, MCP tools, approvals, cost ceilings, data privacy and which models we use.
13 questionsScheduled jobs that check their own output, alerts that reach a person, runbooks, cost dashboards and drift checks.
9 questionsOctelion Data & AI Solutions is a small engineering company based in Bengaluru, India. We build data platforms, reports, AI assistants and agents, automated workflows, monitoring, and the web apps that put all of that in front of a team. The work is delivered as a running system with documentation, not as a notebook or a slide deck.
Operations, finance, product and analytics teams at small and mid-sized companies, mostly in the United States and India. The usual starting point is one of three problems: a report nobody trusts or that takes too long to open, a data platform that is expensive or fragile, or an AI idea that works as a demo and cannot be put in front of real users.
Four things. We hand over running systems, not prototypes. We put one number in writing for the whole scope before work starts. We show our work on public data so anyone can check it. And we build the parts most projects skip: safety checks, review queues, monitoring, audit trails, runbooks and cost controls.
Send one email to hello@octelion.dev with three lines: what is slow, missing or risky; what you have now; when you need it. We reply within one working day with questions, then propose a thirty to forty-five minute scoping call. There is no form and no calendar link on purpose.
We do not publish rates. After the scoping call you get a written number for the whole scope, with what is in and what is out, and that number is fixed unless the scope changes. Engagements range from a few days for a single fix to a few months for a full platform or product. How we work describes the process step by step.
We do not do staff augmentation or hourly bodies on a contract. We do not build on top of client data we are not allowed to see, and we do not put an AI system in front of real users without a safety layer, evals and a person who can review it. We do not publish client names or client numbers; every public example on this site runs on public datasets.
Because a claim you cannot check is a slogan. The home page answers questions over 4.3 million New York taxi trips from June 2025, computed with DuckDB from the public file, and the queries are published in the lakehouse walkthrough. The Power BI walkthrough uses Microsoft's Contoso sample. Client work stays private; the methods are shown in the open.
Databricks on Azure with Delta Lake and Unity Catalog for data platforms. Power BI for reporting, including Fabric where a client is already on it. Claude and other frontier models for AI assistants and agents, with MCP where a standard tool interface helps. Python and Power Automate for workflows. Next.js, Supabase, Vercel and Azure for web apps. DuckDB for fast analysis of files. We pick within that set for the problem, not for the invoice.
No. We work remotely from Bengaluru, with working hours that overlap US evenings and mornings and the full Indian working day. Demos happen weekly on a call; everything else runs in writing and in your repositories.
Send it. If it is one others will have, it joins this page with the answer.
