Project ideas from Hacker News discussions.

Show HN: I implemented a neural network in SQL

Original Article

Hacker News Discussion

📝 Discussion Summary (Click to expand)

Key Themes from the discussion

Theme	Summary	Representative Quote
1. excitement about “neural networks in SQL”	Several users expressed surprise and admiration after seeing neural‑net workloads expressed in SQL, calling the idea impressive despite the steep learning curve.	`"initially rolled my eyes at "neural networks in SQL," but after reading the code I came away impressed"` – 0xnyn
2. relational algebra as a natural IR for tensor programs	The conversation highlighted that relational algebra can serve as an intermediate representation (IR) that database optimizers can reason over, making SQL a potentially ideal language for writing tensor (neural‑net) programs.	`"SQL, as a data oriented and logic PL, might be ideal for writing tensor programs."` – alxmrs
3. broader context & historical parallels	Users placed the idea in a longer tradition of embedding complex computations in relational systems (e.g., DOOM in SQLite, MADlib, and research on einsum‑join equivalence), noting both novelty and precedent.	`"Mathematically, einsum and database joins are the same thing, just over different semirings (real numbers for einsum, booleans for databases)."` – sporkl

All quotations are reproduced verbatim with double‑quote delimiters and credited to their respective usernames.

🚀 Project Ideas

Generating project ideas…

TensorSQL

Summary

A lightweight library that lets users express tensor computations in readable SQL‑like syntax, automatically translated to optimized einsum/tensor operations.
Core value proposition: democratize high‑level tensor programming for data scientists and database users.

Details

Key	Value
Target Audience	Data scientists, ML engineers, database developers
Core Feature	Declarative tensor query language compiled to einsum
Tech Stack	Python, ANTLR4 SQL parser, NumPy/JAX, FastAPI
Difficulty	Medium
Monetization	Revenue-ready: SaaS subscription $12/mo

Notes

HN commenters lamented “just going back to making my CRUD endpoints” and wanted easier ways to write tensor programs; this tool directly addresses that frustration.
Potential for discussion around integrating with existing SQL ecosystems and lowering the barrier to tensor‑based analytics.

EinsumDB

Summary

A command‑line extension that adds einsum‑style tensor functions to PostgreSQL and SQLite, enabling in‑database multi‑dimensional array queries.
Core value proposition: let users run tensor algebra natively inside existing relational databases without data movement.

Details

Key	Value
Target Audience	Database engineers, research scientists, data warehouse admins
Core Feature	SQL‑level tensor ops with automatic broadcasting and differentiation hooks
Tech Stack	PostgreSQL C extension, Rust, ONNX runtime, SQLite VTab API
Difficulty	High
Monetization	Hobby

Notes

Users like “using relational algebra as the IR, letting a database optimizer reason about tensor programs” and wonder about ISO/IEC 9075‑15 MDA support; this tool ships that capability.
Sparks conversation about extending SQL standards and could lead to collaborative standardization efforts.

RelTensorIDE

Summary

A browser‑based IDE and cloud service that visualizes and compiles relational algebra scripts into tensor programs, targeting end‑to‑end model pipelines.
Core value proposition: simplify the creation, testing, and deployment of tensor‑centric ML workflows using relational constructs.

Details

Key	Value
Target Audience	ML engineers, data scientists, startup dev teams
Core Feature	Visual relational algebra builder that generates executable tensor pipelines for Snowflake/BigQuery
Tech Stack	TypeScript, React, Dynabus (Dyna‑inspired), Cloud Functions
Difficulty	Medium
Monetization	Revenue-ready: Freemium with pay‑per‑compute $0.01 per 1k ops

Notes

Echoes “I would have had the same gut reaction… I think that SQL, as a data oriented and logic PL, might be ideal for writing tensor programs” – users will love a UI that makes that possible.
Provides a practical platform for the kind of experimental “SQL‑tensor” demos discussed on HN, encouraging community extensions.