Show HN: I built AI that turns 4 hours of financial analysis into 30 seconds

15•superproton•5mo ago

I built Duebase AI to solve a problem I kept running into in fintech - analyzing UK company financial health takes forever. The process usually goes: download PDFs from Companies House → manually extract data to spreadsheets → calculate ratios → interpret trends. Takes 3-4 hours per company and requires serious financial expertise. The technical challenge: Companies House filings are messy. Inconsistent formats, complex accounting structures, missing data, and you need to understand UK accounting standards to make sense of it all. My approach:

Parse 15M+ UK company records from Companies House API Built ML models to extract and normalize financial data from varied filing formats Created scoring algorithms that weight liquidity, profitability, leverage, and growth trends Generate 1-5 health scores with explanations in plain English

What it does:

Instant financial analysis of any UK company (30 seconds vs 4 hours) Real-time monitoring with alerts for new filings/director changes Risk detection that catches declining trends early No financial background needed to understand results

The hardest part was handling the data inconsistencies - UK companies file in different formats, use various accounting frameworks, and often have incomplete information. Had to build a lot of data cleaning and normalization logic. Currently focused on the UK market since I know the regulatory landscape well, but the approach could work for other countries with similar public filing systems. Link: https://duebase.com

Comments

simonklitj•5mo ago

Any stats on hallucinations or just general inaccuracies in the final results it generates?

nehal3m•5mo ago

Seconded. Throwing a bunch of data into an LLM and calling the result financial analysis is one thing, the results being at all trustworthy and valuable is another. I’d like to see an independent piece of analysis compared to this bot’s output for the same company and the original data.

noduerme•5mo ago

On an even more basic level, throwing data at an LLM and telling people you "built an AI" pretty much blows any chance of anyone taking the results seriously.

potamic•5mo ago

Aren't there services that provide normalized financial data? I don't know about the UK, but I've seen in other markets. If your USP is more towards analytics than data wrangling, it might save you a lot of headache to offload that to someone else.

superproton•5mo ago

Good point. There are a few (Creditsafe, Experian) but they're expensive and focus on basic credit metrics, not comprehensive financial health analysis. The real value is in the contextual interpretation - understanding what the numbers mean for investment/credit decisions. That requires domain knowledge baked into the analytics, not just clean data feeds. Plus controlling the full pipeline lets us iterate quickly on scoring algorithms based on user feedback. Hard to do that with third-party normalised data.

potamic•5mo ago

If these are already expensive, there must be a lot of value for the data itself? Why not make that your core offering? If you're able to automate the process and provide more data on top, you'll definitely have an edge over them.

gitmagic•5mo ago

From the footer:

“Instantly create your dream vacation with Duebase AI, where AI-driven precision meets personal preferences. Tailor-made travel plans crafted just for you, effortlessly.”

Did you pivot?

nehal3m•5mo ago

Looks like OP still does that: https://news.ycombinator.com/submitted?id=superproton

If we want to be charitable their previous website may have served as a template.

gitmagic•5mo ago

Well, at least this one made it to the front page!

sergiomattei•5mo ago

> Our AI analyzes balance sheets, cash flows, and profitability metrics to deliver executive summaries, risk assessments, and key insights that help you make confident business decisions.RetryClaude can make mistakes. Please double-check responses.

Slight issue on the homepage ;)

doppp•5mo ago

#vibecoding

johnisgood•5mo ago

Hey I had an example URL in a README.md with github.com/yourusername. I think this is going to bite me in the ass if anyone checks the commit that fixes it.

incone123•5mo ago

Recommend you change this footer text: Instantly create your dream vacation with Duebase AI, where AI-driven precision meets personal preferences. Tailor-made travel plans crafted just for you, effortlessly.

superproton•5mo ago

Thanks!!!

caretoelaborate•5mo ago

Also under the heading “Comprehensive Accounts Analysis” you have “RetryClaude can make mistakes. Please double-check responses.”

incone123•5mo ago

And I'm curious: was this a pivot from another industry, or did you just try out Claude on a dummy project?

superproton•5mo ago

I use Claude to help me with the copy, I found is the best LLM for that, at least work for me.

latexr•5mo ago

> I use Claude to help me with the copy

“Help with” implies you do some of the work yourself. But most of the comments in this thread so far are about glaring copy mistakes. Have you not read any of the text Claude gave you? That doesn‘t inspire any confidence that your products is remotely competent and what you claim.

dude250711•5mo ago

Thank you Claude, good job!

jsmith99•5mo ago

UK company filings are already marked up with machine readable, structured, standardised tags. It's called XBRL. So it doesn't necessarily need a LLM to parse.

superproton•5mo ago

Wrong, my friend, that is only for small companies, big companies accounts are only available in PDF. Plus, in any case you have the raw data, here you have a full financial analysis.

jsmith99•5mo ago

Large companies need to do it too, including all LSE listed. Here is a summary by the regulator https://www.frc.org.uk/library/digital-reporting/structured-...

But of course a LLM could theoretically automate much of the analysis stage.

tovej•5mo ago

There's a copy pasted "Retry Claude" sentence in the copy of the page, the footer advertises unrelated travel services. Oh, and there's no details on what methods the AI uses for analysis, or any measurements at all of an unproven AI analysis tool.

This is pretty bottom of the barrel.

johnnienaked•5mo ago

Using an AI for this is like using crypto to pay for something: only done for the novelty

Regular coding, the SECs API (or UK equivalent), and some simple spreadsheet formulas have already done the brunt of this for years now.

superproton•5mo ago

Fair point - I started with the Companies House API and traditional parsing too. The issue isn't data access, it's consistency. UK companies file in wildly different formats - some use full GAAP, others micro-entity accounts, many have incomplete data. Rule-based systems work for ~60% of companies, then break on edge cases. Financial health also requires context: Is this debt ratio concerning for their industry/size? Are these cash patterns seasonal or declining? How do you weight profitability vs growth stage? I spent months on traditional approaches first. The AI handles the inconsistency and contextual interpretation that spreadsheet formulas can't.

johnnienaked•5mo ago

I respect your comment but all of these things can be automated without an AI

The inconsistency of UK reporting is highly interesting though

zarzavat•5mo ago

The heading text is cut off by the top navigation (on mobile safari).

monkeydust•5mo ago

I use Companies House (CH) a fair bit so interesting, tried out:

- There is UX issue here as on first pass it just looked like a view to CH, the scores where N/A, after a minute it came to life with something useful but during that period wasn't sure what was going on, you need to manage the user and let them know its working out scores not show N/A.

- Liked the history of scores as well, perhaps a chart would be good here.

- I need to understand how you score before I would ever consider paying for this, there has to be some explanation of the methodology.

superproton•5mo ago

That's great feedback, thanks a lot!

South Korean crypto firm accidentally sends $44B in bitcoins to users

Apache Poison Fountain

Web.whatsapp.com appears to be having issues syncing and sending messages

Google in Your Terminal

Shannon: Claude Code for Pen Testing

Anthropic: Latest Claude model finds more than 500 vulnerabilities

Brooklyn cemetery plans human composting option, stirring interest and debate

Why the 'Strivers' Are Right

Brain Dumps as a Literary Form

Agentic Coding and the Problem of Oracles

Malicious packages for dYdX cryptocurrency exchange empties user wallets

Show HN: I built a <400ms latency voice agent that runs on a 4gb vram GTX 1650"

Penisgate erupts at Olympics; scandal exposes risks of bulking your bulge

Arcan Explained: A browser for different webs

What did we learn from the AI Village in 2025?

An open replacement for the IBM 3174 Establishment Controller

The P in PGP isn't for pain: encrypting emails in the browser

Show HN: Mirror Parliament where users vote on top of politicians and draft laws

Ask HN: Opus 4.6 ignoring instructions, how to use 4.5 in Claude Code instead?

We Mourn Our Craft

Jim Fan calls pixels the ultimate motor controller

Exploring a Modern SMTPE 2110 Broadcast Truck with My Dad

AI UX Playground: Real-world examples of AI interaction design

The Field Guide to Design Futures

The Other Leverage in Software and AI

AUR malware scanner written in Rust

Free FFmpeg API [video]

Are AI agents ready for the workplace? A new benchmark raises doubts

Show HN: AI Watermark and Stego Scanner

Clarity vs. complexity: the invisible work of subtraction

South Korean crypto firm accidentally sends $44B in bitcoins to users

Apache Poison Fountain

Web.whatsapp.com appears to be having issues syncing and sending messages

Google in Your Terminal

Shannon: Claude Code for Pen Testing

Anthropic: Latest Claude model finds more than 500 vulnerabilities

Brooklyn cemetery plans human composting option, stirring interest and debate

Why the 'Strivers' Are Right

Brain Dumps as a Literary Form

Agentic Coding and the Problem of Oracles

Malicious packages for dYdX cryptocurrency exchange empties user wallets

Show HN: I built a <400ms latency voice agent that runs on a 4gb vram GTX 1650"

Penisgate erupts at Olympics; scandal exposes risks of bulking your bulge

Arcan Explained: A browser for different webs

What did we learn from the AI Village in 2025?

An open replacement for the IBM 3174 Establishment Controller

The P in PGP isn't for pain: encrypting emails in the browser

Show HN: Mirror Parliament where users vote on top of politicians and draft laws

Ask HN: Opus 4.6 ignoring instructions, how to use 4.5 in Claude Code instead?

We Mourn Our Craft

Jim Fan calls pixels the ultimate motor controller

Exploring a Modern SMTPE 2110 Broadcast Truck with My Dad

AI UX Playground: Real-world examples of AI interaction design

The Field Guide to Design Futures

The Other Leverage in Software and AI

AUR malware scanner written in Rust

Free FFmpeg API [video]

Are AI agents ready for the workplace? A new benchmark raises doubts

Show HN: AI Watermark and Stego Scanner

Clarity vs. complexity: the invisible work of subtraction

Show HN: I built AI that turns 4 hours of financial analysis into 30 seconds

Comments