Built for the Databricks Community Genie-Powered App Challenge 2026 — Track A: Real-World Problem Solver
THE PROBLEM THAT DOESN'T SHOW UP ON A DASHBOARD
Most small businesses don't fail because they run out of customers. They fail because they discover something is wrong two weeks too late — after a commitment is already due, after a supplier has already called, after the point where a warning would have actually mattered.
The tools meant to prevent this are usually dashboards: rows of KPIs, a chart of daily activity, maybe a metric buried in a report nobody opens on a Tuesday. Dashboards answer questions someone thought of in advance. But the questions a business owner actually carries around in their head are messier than that:
- Can I cover my team's obligations this month?
- Who's been slow to follow through on what they owe us?
- Are we heading into trouble over the next two months?
- What do I need to reorder this week?
- What actually caused that bad month we had?
None of those map cleanly onto a chart. They map onto a conversation — and that's the gap Ledgerly is built to close.
WHAT LEDGERLY IS
Ledgerly is an operations copilot for small businesses, built end-to-end on Databricks and powered by Genie. Instead of asking an owner to interpret a dashboard, it lets them ask questions in plain English and get answers grounded in real data — sales activity, customer accounts, inventory, and a forward-looking outlook on the state of the business.
To keep the project concrete, I modeled it around a fictional but realistic business: Brew & Bloom, a café that also runs a small wholesale coffee operation. The synthetic dataset behaves the way a real business does — daily retail activity, wholesale customers who each work on their own agreed schedule, accounts that don't always get settled on time, seasonal demand swings, inventory that needs reordering, and a deliberately buried rough patch. If Ledgerly can catch and explain that rough patch, it can do the same for a real business.
GENIE IS THE PRODUCT, NOT A FEATURE BOLTED ON
It would've been easy to build this as a dashboard with a chatbot stapled to the side — which is exactly the pattern that already fails business owners. So Genie sits at the center of the experience, not on top of it as an afterthought.
There's still a lightweight overview panel — quick KPIs and a 90-day outlook — but it's there purely for orientation, a glance-and-go layer. The real analytical work happens inside Genie, where an owner can go past what is happening and actually ask why.
To make Genie reason like an analyst rather than answer like a query tool, it isn't just pointed at tables — it's handed the business logic an analyst would actually use:
Operating buffer — how much room the business has, based on its recent 30-day usage rate At-risk account — an account running more than 15 days past its own agreed schedule, not a generic global cutoff Reorder point — stock at or below its threshold Outlook risk — pulled directly from the 90-day forward-looking model
That distinction is the whole point. A generic text-to-SQL layer can answer "what did we sell." It has no idea what "late" means for one specific account, or what "at risk" means for one specific business. Encoding those definitions is what turns Genie from a query interface into something that reasons the way an owner or bookkeeper actually would. Remove Genie from Ledgerly and you're left with a static dashboard. Leave it in, and Ledgerly becomes something an owner can talk to.
THE ARCHITECTURE
Everything above depends on a clean data foundation underneath it. Ledgerly follows a deliberate lakehouse pipeline, end to end:
Synthetic data
→ Unity Catalog Volume
→ Bronze (raw Delta tables)
→ Silver (cleaned tables)
→ Gold (business-ready tables)
→ 90-day forward-looking forecast
→ Genie Space
→ Databricks App
Synthetic data for Brew & Bloom lands in a Unity Catalog Volume, then moves through the classic medallion layers — raw Bronze, cleaned Silver, and Gold tables shaped specifically for the questions an owner would ask: daily activity, account reliability patterns, inventory levels. From Gold, a 90-day outlook is generated, and that outlook — alongside the rest of the Gold layer — is exactly what Genie is grounded in. The whole thing ships wrapped in a Databricks App, so it's something an owner could genuinely click into, not a notebook that stays on a data team's laptop.
WHY THIS MATTERS BEYOND THE CHALLENGE
The Genie-Powered App Challenge asked for real-world problem solving, and this kind of blind spot is about as real as small-business problems get. It isn't glamorous, but a well-timed answer to "can I cover this month" is worth more than almost any other feature you could ship.
The biggest lesson from building Ledgerly: most of the "AI" value didn't come from anything exotic in the model. It came from precisely defining what the business terms actually meant. Genie is powerful because it reasons over data conversationally — it's useful because it was told, exactly, what "late," "at risk," and "buffer" mean for this kind of business. That's where I'd tell anyone building something similar to spend their time.
Repo: github.com/srimonishan/ledgerly-genie
Contact me: srimonishan.com
Ledgerly was built for the Databricks Community Genie-Powered App Challenge 2026, Track A — Real-World Problem Solver.
monishan
Advika
