LiDAR & 3D perception data
Point-cloud processing, multi-sensor fusion, calibration, ground-plane alignment and auto-labeling. A niche where very few engineers operate, and where I've spent years in production.
open to opportunities & collaborations
Sohail Saifi.
I gravitate toward the problems most engineers avoid. That habit has taken me from machine learning data infrastructure and 3D perception systems to complete products that I design, build, and ship solo.
Whatever the stack, I take ideas all the way to production.
01
What I do
Four disciplines, one thread: hard problems turned into systems people rely on every day.
Data pipelines for ML
End-to-end pipelines that transform raw, chaotic sensor streams into clean, validated, ML-ready datasets, with the throughput, QA gates and automation to run every day without babysitting.
Web platforms & 3D visualization
React / Next.js / Three.js applications that make heavy data explorable: interactive point-cloud visualizers, internal tools and dashboards people rely on daily.
Backends & automation
Production Python backends (Django, FastAPI, Flask), AWS infrastructure, and the scripts and services that quietly remove hundreds of hours of manual work.
02
Selected work
Production systems I've architected and built, for my employer and as products of my own. The proprietary ones I'm happy to walk through in depth on a call.
StitchConsole: embroidery digitizing in the browser
A professional embroidery digitizing platform I designed and built solo, as a free, cross-platform alternative to Wilcom. Hand-crafted stitch-generation algorithms (satin columns with underlay and pull compensation, tatami, contour and spiral fills, auto-digitizing with boundary tracing and color quantization, 2-opt stitch ordering), plus a full vector editor, 3D thread preview, and export to every major machine format (DST, PES, JEF, VP3, EXP).
Try it live at stitchconsole.comLiDAR data-ops pipeline
End-to-end automation of the LiDAR data lifecycle: collection from live sensors, calibration, pre-processing and labeling. What used to be a manual, error-prone process now runs as a pipeline, cutting manual effort by 80% and making dataset production repeatable at scale.
Web-based 3D point-cloud visualizer
An interactive, browser-based visualizer for massive 3D LiDAR scenes: frame-by-frame playback, label overlays and inspection tooling. Turned raw point clouds from opaque binary blobs into something the whole team can open, inspect and debug in seconds.
3D perception algorithm suite
Designed and implemented the geometric algorithms at the heart of the pipeline: automatic ground-plane alignment, multi-sensor point-cloud fusion, label interpolation across frames, and background detection in 3D space. These are the pieces that make auto-labeling actually work.
ML-ready data processing
The bridge between field recordings and model training: streamlined processing of raw multi-sensor data into normalized, validated, ML-ready formats, so perception models train on data that's consistent, complete and trusted.
03
Experience
Current
Mar 2022 – Present
Software Development Engineer · CronAI
- Own the end-to-end LiDAR data platform, from live sensor collection and calibration to automated labeling, cutting manual effort by 80% and turning dataset production into a repeatable pipeline
- Built a full 3D data labeling tool single-handedly, from scratch, replacing the third-party software the company had been paying for every month and bringing the entire annotation workflow in house
- Designed the 3D algorithms at the core of the auto-labeling stack: automatic ground-plane alignment, multi-sensor point-cloud fusion, label interpolation across frames, and background detection
- Built the browser-based 3D visualization tools the whole team relies on daily to open, inspect and debug massive point-cloud scenes
- Turn raw multi-sensor recordings into validated, ML-ready datasets that perception models train on, with QA gates that keep bad data out
- Operate across the full stack: Python backends, AWS infrastructure, and the automation that keeps the pipeline running unattended
Python · TypeScript · NumPy · Pandas · Django · FastAPI · Three.js · WebGL · React · MySQL · AWS (EC2, S3) · Docker · Linux · Shell scripting
Nov 2021 – Mar 2022
Software Development Intern · VectoScalar
- Shipped production features in Python and JavaScript across multiple client projects, going from first ticket to deployed code within the first month
- Designed and integrated REST APIs and backend logic powering client web applications
- Wrote Python scripts that automated repetitive development and testing workflows, saving the team hours every week
- Left with production-grade habits: clean commits, tight code reviews, and delivering under real deadlines
Python · JavaScript · REST APIs · Git
Jun 2021 – Jul 2021
Cyber Security Summer Intern · Gurugram Police
- Researched cybersecurity weaknesses and threat vectors alongside industry experts
- Contributed to hardening security measures and incident-response readiness
Security research · Threat analysis
2018 – 2022
B.Tech, Computer Science & Engineering · ABES Institute of Technology
Where the fundamentals were built, and the obsession with building things started paying rent.
04
Toolbox
Depth where it matters, breadth where it helps.
Data & 3D
- Point-cloud processing
- LiDAR data pipelines
- Sensor fusion & calibration
- 3D geometry & math
- NumPy · Pandas
- Data labeling & QA
Languages
- Python
- TypeScript / JavaScript
- C#
- SQL
- Shell / Bash
Web
- React · Next.js
- Three.js / WebGL
- Node.js · Express
- HTML · CSS · SASS
- UI/UX · Figma
Backend & Infra
- Django · FastAPI · Flask
- AWS (EC2, S3)
- Docker
- MySQL · Firebase
- Git · CI · Linux
05
About
I'm Sohail, a software engineer who ended up in one of the most interesting corners of the industry: teaching machines to make sense of the physical world. I've been glued to computers since childhood, and for the past 5+ years I've turned that obsession into production systems.
My sweet spot is the unglamorous, high-leverage layer everyone depends on but few can build: the data infrastructure underneath machine learning. When the data is 3D, massive and messy, that's when it gets fun. Outside of work I ship products of my own, like StitchConsole, a browser-based embroidery digitizing studio. I care about the details, I automate everything that deserves automating, and I like building tools people actually enjoy using.
06 / contact
Got a hard problem?
A role, a project, or just a good technical conversation. My inbox is open.
sohailxsaifi@gmail.com