Golden Analytics
Golden Analytics is an AI-native BI platform built for data teams who are tired of the tradeoffs in today's tools -- the depth of self-service analytics tools without the rigidity, the accessibility of a modern design tool, and AI that actually augments how analysts work rather than getting in the way.
Open Opportunities
Available Challenges
Golden Analytics — Query Challenge
The Challenge This is a challenge for potential query engineers to solve. The context of the challenge is a business intelligence application. This application will need to issue analytical queries (e.g. selection/projection, aggregation, filtering, joins, calculations) against a set of tables on a remote database server and return the results. The long-term goal would be to support many different relational databases, with this API serving as the translation layer between the application and those databases. The goal of the project is to make an API that can express these kinds of queries and execute them across a set of relational database backends, without having to worry about the details of how they are implemented on each backend. You'll demonstrate this by writing a driver application that uses your API to execute queries against both backends, and a testing framework that demonstrates how you'd test your API against these databases, architected to expand to others in the future. From a language and technology perspective, we would prefer if you demonstrated all of this in Node/TypeScript (this is our current platform), but you can select another platform if you like. You should use the Docker image goldenbeers/golden-query-challenge:1.0 on the project. It contains two databases (MySQL 8 and Postgres 17) pre-loaded with sample data. We expect you to use AI in completing this challenge. Please keep track of your major prompts and how you guided the AI during the project. We also don't expect you to get everything to polished completion. Expect to spend about 45 minutes on the challenge, and keep track of what you would do if you had extra time to spend on it. AI will likely get you most of the way there. After you have your basic API in place, pick one way to extend it (an advanced SQL feature, an interesting third database, etc). Things you should deliver One video (~6 minutes). Start with a short intro: tell us who you are, one thing about yourself that isn't on your resume, and what specifically excites you about Golden Analytics. Then walk us through your API and why you designed it this way. One take is fine. We're listening to what you say, not how polished you sound. A driver application demonstrating some of the queries below and your one extension A test suite for your API using the provided databases A README, written by you (not the AI): summarizing the design of your API; describing your major prompts and where you redirected the AI; what you would do if you had more time Database schema The sample data is the Northwind database (schema diagram). The link shows a schema diagram with the default relationships between tables. Sample queries The most recent N orders The count of customers from a given city, e.g. 'london', or 'mexico d.f.' Average number of units per order For each employee, the number of shipped vs. unshipped orders The total freight / total order price ratio per customer The 75th percentile of order price by customer's country
Household Portfolio Rebalancer
Description Someone manages their household's investments across multiple broker accounts and wants to keep their portfolio aligned to a chosen asset-allocation target — and to be able to change that target over time. Their broker only gives them a flat CSV export of raw positions (symbol-level holdings), not a view organized the way they actually think about their money: File to be found here. Your job is to close four gaps: The data isn't in a usable shape. The provided CSV is a flat, symbol-level export across multiple accounts. Turn it into something organized the way the user actually thinks about their money. There's no concept of a target. The user thinks in terms of asset classes (US Equity, International, Gold, Cash, Treasuries, etc.) and target percentages, but the brokerage only shows individual ticker holdings. You'll need to design how tickers map to asset classes — the sample data does not come with this mapping; that design is part of the challenge. Rebalancing math is tedious and error-prone by hand. Given a current allocation and a target allocation, figure out exactly which symbols to buy and sell — and how much — to reach the target. Cash cannot move between accounts, so each account must be rebalanced independently, funded and absorbed only by its own money-market/cash position. Some accounts are more liquid than others. The user may prefer to hold more cash in certain accounts (e.g., maximize cash in a brokerage account rather than a retirement account) because that account is more accessible. Your solution should account for this preference. Build a working tool — not a script or notebook — where a user can see their current allocation, edit a target allocation, and get back the exact set of transactions needed to reach it. Constraints to Consider No paid or proprietary services required. Your solution must run locally from documented setup steps — no infrastructure budget assumed. The CSV export format is fixed. You cannot change what the brokerage sends. Your solution must ingest the provided format as-is. Cash cannot move between accounts. Each account rebalances independently, funded only by its own money-market/cash position — never assume you can use one account's cash to fund a trade in another. 120-minute total budget, including your video. Prioritize a complete, working core experience over a partially-built comprehensive one. AI Usage Guidance We expect you to use AI tools while building this. We evaluate how you use them — not whether you used them. Evidence of iteration, redirection, and critical evaluation scores higher than a polished output with no process documentation. Note: this challenge does not require your solution to call an AI API at runtime — the AI evaluation is entirely about your build process. The single highest-signal indicator: your video answer to the mandatory AI question. If you cannot name a specific moment where you redirected AI output, evaluators will assume you did not. Mandatory AI question for your video: Walk me through one moment where you disagreed with, pushed back on, or redirected what the AI gave you — and what you did instead. Name the specific moment. Explain what the AI produced that didn't meet the bar, what you did differently, and why. Speak naturally. Communication is assessed on clarity of technical ideas and logical structure — not verbal polish, accent, or filler words. Submission: Upload each deliverable as a separate file directly on the Provn platform: your primary artifact, your README document (Sections A, B, and C), and your video walkthrough (MP4 or MOV).
Turn Business Data Into Answers
The Vision Most business users can't access their own data. The marketing analyst who needs last month's regional numbers has to file a ticket and wait two days. The ops manager tracking deal velocity is working from a spreadsheet that's three weeks old. The business owner who wants to understand their top customers has no idea where to even start. Golden Analytics is building toward "Canva for data" — a world where any business user can understand, explore, and act on their data, without needing an engineer, a BI tool license, or a SQL degree. The people we're building for are not technical. They are smart, they have real questions, and they are currently stuck. You're joining the team. Your first challenge is also the most important one we work on every day: how do you make data genuinely useful for someone who has never written a query? The Challenge You have 30 minutes and a real dataset to work with: Washington State fiscal data (link to data set below as well) — public government spending broken down by fund type and fiscal year. This is real data, with real shape and real quirks. You do not need a database connection; you can embed the data directly or load it from the file. Your task: Propose one way to help a non-technical user get value from this data. Then build it as a proof-of-concept web app. The solution is yours to design. We want to see your product instincts alongside your engineering. What do you think would genuinely help someone who has never looked at a government budget? A city councilmember trying to understand where the money went. A journalist tracking spending trends. A policy analyst who knows the questions but not the SQL. Build something for one of them. Constraints Every solution — whatever direction you take — must satisfy these two things: It must target a non-technical user. The person using your app has never written a query. Design for them. SQL must not be visible unless the user has explicitly requested it. It must be a working web app. A mock or stub is acceptable for data that isn't wired up, but the UI must be functional and the core interaction must work. If you can demo it, even better. AI usage: Golden Analytics builds AI-native features — it's a core part of what this role does. If your solution includes an AI or intelligent component, great. If it doesn't, explain in your README why you made that choice and where you see AI adding value in a future iteration. Either approach is valid. What we're looking for is intentional product thinking, not checkbox compliance. If you use AI/ML components: All inputs to any model, vector store, or intelligent component must be logged — even in the POC. You do not need a persistent log store, but your code must show clearly where and how logging would happen. This is a governance requirement in enterprise B2B contexts. This is a proof-of-concept. You are not expected to ship production code in 30 minutes. What we're evaluating is your thinking, your trade-offs, and how you approach the problem — not polish. AI Usage Guidance We expect you to use AI tools. We evaluate how you use them — not whether you use them. Evidence of iteration, redirection, and critical evaluation scores higher than a polished output with no process documentation. The single highest-signal indicator: your video answer to the mandatory AI question. If you cannot name a specific moment where you redirected AI output, evaluators will assume you did not. Mandatory AI question (answer in your video): Walk me through one moment where you disagreed with, pushed back on, or redirected what the AI gave you — and what you did instead. Name the specific moment. Explain what the AI produced that didn't meet the bar, what you did differently, and why. Dataset Vendor-Payments\2021-23.xlsx