How do you turn raw scan data from a physical device into a browser-based 3D building model your team can query, annotate, and collaborate around?
Azati provided 24 months of full-cycle engineering for a US construction tech startup's SaaS + IoT digital twin platform: browser-based 3D viewer, cloud processing pipeline for large point cloud datasets, backend microservices, real-time collaboration, and AWS infrastructure. industry: Construction & PropTech
months of full-cycle product engineering, from MVP features to production infrastructure
3D rendering engines integrated in one platform: Three.js, Autodesk APS Viewer, and Potree
scan-to-view workflow delivered: hardware device to browser-based 3D digital twin
A US construction tech startup needed engineering capacity across the full stack of their SaaS + IoT digital twin platform: the browser-based 3D viewer, cloud pipeline for large point cloud datasets, backend microservices, real-time collaboration, and AWS infrastructure, all working as one coherent product.
Azati provided that capacity for 24 months, contributing to every layer: from the 3D viewer built with Autodesk APS Viewer SDK, Potree, and Three.js, to the AWS Batch processing pipeline, NestJS microservices, and Terraform-managed infrastructure. The result was a working scan-to-view workflow: scan a building, process the data in the cloud, view and collaborate on the digital twin in a browser.
The platform needed to handle CAD/BIM-style model viewing, raw point cloud visualization from the scanning device, and custom interactive geometry in the same browser-based experience. Each required a different rendering approach and a different SDK:
Raw scan output is large, computationally heavy, and needs significant processing before it becomes a viewable digital twin. Building a reliable, scalable pipeline to handle this asynchronously and at variable volumes was a core engineering challenge:
The platform started with manually managed virtual machines, and migrating to a reproducible, automated infrastructure while the product was in active development required care and a clear incremental path:
The scanning hardware introduced a dimension that pure SaaS products don't have: the engineer contributing to the platform also supported requirements for the device firmware side, bridging the gap between what the hardware produced and what the software expected:
This engagement required someone who could work across Angular frontend, NestJS backend, AWS cloud infrastructure, 3D visualization SDKs, and DevOps, all as part of one coherent product. The breadth of full-cycle engineering capability matched what a startup at this stage actually needed.
Combining a CAD/BIM-style model viewer, a point cloud renderer, and custom 3D geometry in one browser-based product requires a depth of computer graphics knowledge that most web developers don't have. That specific background was a primary reason Azati was chosen for this engagement.
Digital twin platforms for buildings sit at the intersection of computer graphics, cloud infrastructure, and IoT hardware in ways that no tutorial covers end to end. Many of the problems that came up during the engagement had no standard answers, and the ability to work through them without escalating or blocking was explicitly cited as a reason Azati was brought in.
Azati handled the full infrastructure migration from manual VMs to Terraform-managed AWS, covering VPC, ECS, S3, SQS, Redis, and CloudWatch, as part of the same engagement, not as a separate workstream.
Browser-based 3D viewers, large dataset processing pipelines, real-time collaboration, and AWS infrastructure are all things Azati has built together in one product. Let's talk about yours.
Launch my 3D SaaS productThe platform required engineering work across six distinct technical areas, all of which had to work as an integrated product. Azati contributed to all of them.
The core user-facing component of the platform, the 3D viewer, was built using three different rendering technologies depending on the data type. Autodesk Forge/APS Viewer SDK with custom extensions handled structured CAD/BIM-style models. Potree handled the large, unstructured point cloud data from the scanning device. Three.js powered custom interactive geometry, overlays, and visualization features built directly into the browser.
Raw scan data is large and computationally heavy. An AWS Batch and SQS-based orchestrator queues and processes incoming scan datasets asynchronously, handling the parallel processing needed to turn raw device output into viewable digital twin formats. S3 stores both raw and processed data, and CloudWatch provides telemetry across the pipeline.
NestJS and TypeScript microservices handled the platform's business logic, data management, and API layer. Python services contributed to the data processing and AI model components. MongoDB provided flexible document storage, and Redis handled caching and session data.
Two custom AI models were built as part of the platform. A face blurring model automatically anonymizes people captured in building scans, protecting privacy. A point cloud to blueprint conversion model generates 2D floor plan representations from 3D scan data, making the scan output useful for workflows that work better with 2D representations.
Multiple stakeholders needed to work with the same building model simultaneously and be kept informed as processing completed. WebSocket and SSE-based real-time update flows notified users when scans finished processing and became available to view. Slack and Discord integrations extended collaboration into the communication tools teams were already using.
Azati managed the full AWS infrastructure using Terraform for infrastructure-as-code and migrated from manually managed virtual machines to a reproducible, automated setup using cloud-init and Ansible. The infrastructure covered VPC configuration, ECS containerized services, S3, SQS, Redis, and CloudWatch monitoring.
| Area | What was delivered |
|---|---|
| 3D viewer | Autodesk APS Viewer SDK with extensions, Potree point cloud viewer, Three.js geometry |
| Cloud pipeline | AWS Batch + SQS processing orchestrator for large 3D scan datasets |
| Backend | NestJS/TypeScript and Python microservices, MongoDB, Redis |
| AI models | Face blurring model, point cloud to 2D blueprint conversion model |
| Real-time collaboration | WebSocket and SSE-based updates, Slack and Discord integrations |
| Infrastructure | Terraform-managed AWS, VM migration to cloud-init/Ansible, Docker/ECS |
| Monitoring | CloudWatch telemetry across the processing pipeline and services |
| Firmware support | Requirements coordination between the scanning device and the cloud platform |
Azati contributed across all major platform components for 24 months on a Time and Material basis, covering frontend, backend, cloud infrastructure and CI/CD, AI components, integrations, and ongoing product support and optimization, not just one layer of the stack.
The platform's requirements evolved as the startup worked through the complexity of combining hardware, cloud processing, and a SaaS product. Agile delivery gave the engineering work the flexibility to absorb that evolution:
From the physical scanning device through cloud processing to a browser-based 3D digital twin, the full pipeline was built and operational, which is the core product promise of the platform.
Multiple stakeholders: construction teams, property assessors, insurers, could work inside the same building model simultaneously. Slack and Discord integrations brought notifications into the tools teams already used, reducing context switching during active projects.
The migration from manually managed VMs to Terraform-managed AWS with Ansible and cloud-init configuration established a maintainable, automatable infrastructure foundation that could grow with the product.
The engagement left Azati with real depth in browser-based digital twin products: how scan-to-view pipelines behave under real-world dataset sizes, how to extend 3D SDK viewers without breaking performance, and how to bridge hardware device workflows with cloud SaaS development, a combination that rarely exists in a single team.
What this engagement demonstrates:
The gap between "web developer" and "3D viewer developer" is larger than most teams expect. Viewer SDK extensions, point cloud rendering performance, and custom geometry interactions all require knowing how the GPU pipeline works, not just how the browser DOM works. That gap is where most digital twin frontend efforts stall.
Infrastructure, AI models, real-time communication, firmware requirements, and 3D rendering don't come apart cleanly at a startup stage. The value of a full-cycle engineering contribution is that all of it gets handled without the startup having to coordinate across separate specialists for every domain.
When a physical scanning device is at one end of the pipeline and a browser viewer is at the other, the failure modes are different from a pure SaaS product. Data arrives in unpredictable volumes, at unpredictable times, in formats that change as the hardware firmware evolves. Infrastructure that works in a demo breaks in production if it wasn't designed for that reality from the start.
Azati delivered a production-grade real estate catalog platform at 3х speed using AI coding agents with complete senior-engineer ownership from architecture to deployment.
Large-scale ECM system for online property auction marketplace.
ML and CV prototype for automated road defect detection.
Progressive web scraping platform compiling customer data dashboard.
Cross-platform mobile app for on-site inspections with geolocation tracking.
Reporting platform automating equipment depreciation reporting process.
Last updated
