bmadduma

Working on automating small business finance (bookkeeping, reconciliation, basic reporting).

One thing I keep noticing: compared to programming, accounting often looks like the more automatable problem:

It’s rule-based Double entry, charts of accounts, tax rules, materiality thresholds. For most day-to-day transactions you’re not inventing new logic, you’re applying existing rules.

It’s verifiable The books either balance or they don’t. Ledgers either reconcile or they don’t. There’s almost always a “ground truth” to compare against (bank feeds, statements, prior periods).

It’s boring and repetitive Same vendors, same categories, same patterns every month. Humans hate this work. Software loves it.

With accounting, at least at the small-business level, most of the work feels like:

normalize data from banks / cards / invoices

apply deterministic or configurable rules

surface exceptions for human review

run consistency checks and reports

The truly hard parts (tax strategy, edge cases, messy history, talking to authorities) are a smaller fraction of the total hours but require humans. The grind is in the repetitive, rule-based stuff.

I'm telling you from my experience, the idea of “doing things that don’t scale” no longer works for enterprise B2B startups.

Every customer’s data systems and requirements are different. You can build a consulting company that way, but not a scalable product company.

Believe it or not, this comes from my own experience.

Just want to see hackers response about text to sql and RAG. It is promising to generate SQL using GPT4 and run for small databases but it completely fails with databases with hundreds of tables. GPT doesn't know how to generate complex queries even with enough metadata. This is my problem, copilot like enterprise chatbots with RAG is hard to build to get 90% - 95 accuracy Any solutions ?

Buddhika, Kelum, and Chong Han here. We are building a self-hosted data infrastructure platform for computer vision. Our community page is https://www.layernext.ai/community

In the past, we worked on a couple of high-scale computer vision projects in retail, farming, and hospitals in various capacities. These projects involved 2D object sections, 3D object tracking, and more advanced 3D perception. Like other CV Engineers, we observed a common factor during these projects: one needs a large volume of high-quality data to build a production-deployable CV system. Our biggest challenge was not having a robust data infrastructure to handle large volumes of data. Our S3 buckets were like a data swamp; we had so much raw image and video in storage buckets without tracking. Instead of working on CV, we had to develop tools for data operations.

We understand that many of us have our own custom scripts and stitch them together to make things happen in the CV pipeline. However, it is brittle and cumbersome to maintain.

We wanted to build a system on top of the cloud buckets such as S3 that store all file indexes, labels, metadata attributes, inference outputs, model training outcomes, and literally anything related to machine learning/computer vision. This makes it possible for us to search for anything and consume efficiently. This behaves as a DataLake (by the way, "DataLake" is an overused term). All other downstream processes in the CV pipeline can access data more efficiently via SDK and can also return data back to the Lake (e.g., training/inference outcomes).

The reason we made it self-hosted is to address data security and privacy concerns. Since data is fundamental to AI, we believe that companies and organizations should have complete control over it.

Currently, we support AWS, GCP, and Azure cloud buckets; soon, we will support local storage. We ship this as a Docker container so you can just install it on any VM or local server. The installation script will do all the configuration automatically. The Python SDK and documentation are available but not perfect yet.

We’ve launched this under MIT and Elastic licenses so any developer can use it. Our goal is not to charge individual developers. We make money by charging a license fee for things like multiple users, multiple buckets, scalability with K8, and providing support.

Give it a try: https://www.layernext.ai/community Let us know what you think.