← All workCase 05 / 07
Live at buniyadbyte.comCivil-engineering SaaS

Structural drawings in,cut lists out.

A civil-engineering SaaS that reads AutoCAD, PDF and Revit drawings, computes bar-bending schedules with industry formulas, and plans steel cutting to minimise waste.

Product
BuniyadByte buniyadbyte.com↗
Client
Construction tech · India
Role
Full-stack engineer
Surfaces
AI drawing reader · Org panel · Admin
The brief

Before steel reaches site, engineers turn structural drawings into a bar-bending schedule — every bar's shape, length and count — usually by hand, in spreadsheets, for days.

This platform does it from the drawings themselves: import a DWG, PDF or Revit model, apply the codes and formulas, then decide how to cut stock bars so the least steel ends up as scrap.

01

Engineering workspace

Projects, blocks and floors; drawing import; beam, column, slab and footing schedules.

02

Drawing service

A dedicated service that parses DWG, PDF and Revit files into structured members.

03

Organisation portal

Team access per project, subscriptions and add-ons.

04

Admin panel

Customers, shape and bar libraries, pricing and platform analytics.

System map

How it fitstogether.

Every surface talks to one platform core. This is the shape of the system as it runs in production.
Engineering workspaceOrg portalAdmin panelPlatform coreCore API+ drawing serviceDrawing ingestionFormula engineWaste optimiserOffcut inventoryBar-fill geometryOpenAI visionAutodesk Platform ServicesCAD & PDF toolchainRazorpayS3 + CDN
The hard parts

What made ithard — and how.

The problems that don't show up in a screenshot, but decide whether a product survives real users.
01

Cutting steel with the least waste

The problem

Choosing which stock bars to cut for thousands of required lengths is a classic hard optimisation problem — and engineers export reports while others are working.

The approach

An adaptive selector scores each bar diameter's demand and picks a solver — greedy, local search, dynamic programming or branch-and-bound — each with time budgets. They run on a shared worker pool so heavy exports never block the app.

Cutting-stockBranch and boundWorker pool
02

Laps, couplers and reusable offcuts

The problem

Long bars must be spliced, and leftover pieces are real steel worth reusing.

The approach

Long bars are split by effective coverage to count splices and lap length, couplers apply only where the bar size allows, and usable offcuts return to inventory with their lineage for the next run.

Splicing rulesInventory genealogy
03

Formulas as data, not code

The problem

Every bar shape has its own bends and deductions, and codes vary by grade and project.

The approach

Shapes carry their own bend and deduction expressions, evaluated per end, with development length from grade tables or an editable lap coefficient — so new shapes need data, not a deploy.

Rule engineBend deduction
04

Geometry for bar fills

The problem

Slab bars must fill irregular boundaries at an exact spacing.

The approach

Pure planar geometry — point-in-polygon, area and a sweep with bisection — lays parallel bars inside any outline, ported from a desktop CAD prototype and matching it exactly.

Computational geometry
AI in production

An AI that reads drawings — and a system that checks it

Many drawings arrive as PDFs. Vector PDFs go through a deterministic text-layer engine first; scanned or complex sheets fall back to a vision model that works with tools — it can think, zoom into a region, confirm a reading and add a member. The model only reads raw text; code derives diameters and spacing deterministically.

  • Deterministic extraction first, AI only as fallback
  • Tool-using vision agent: think, zoom, confirm, add
  • Page and layout classification
  • Model reads, code calculates — no hallucinated numbers
Inside the system

Everythingit does.

01DWG, PDF, Excel and Revit import with merge and clone
02Beam, column, slab and footing schedules
03Bulk recalculation of cover, laps, grades and hooks
04Continuous beam and slab bar generation
05Waste reports with lap vs coupler options
06In-browser length measurement on CAD models
07Per-project access for organisation teams
08Subscriptions with plan-gated features
Built with
Built with
01

Backend

Node.js · TypeScript · FeathersJS · MongoDB · BullMQ · Piscina

02

CAD & documents

LibreDWG · DXF · MuPDF · PDF.js · OpenCV · Autodesk APS

03

AI

OpenAI vision · Tool calling

04

Web & platform

Next.js · MUI · Razorpay · S3 + CloudFront · CI/CD

Building something like this?

Let’s scope yours.