Available for opportunities

Ronak
Parmar

Systems Builder & Open Source Creator

I build from first principles — compilers, CLIs, AI pipelines, and full-stack systems. Not just code that works, but architecture that lasts.

Core Systems
AI / ML
Infrastructure
Full Stack
§ 01 — Philosophy

How I think
about building.

I started writing code the way most engineers do — copying patterns, assembling libraries. Then I built a compiler from scratch in C11 with no parser generators, no external runtime. That changed how I see software.

Every abstraction is a trade-off. Every layer has a cost. I'm interested in understanding systems at the level where trade-offs become visible — and building things that are intentional about them.

My work spans from low-level systems programming (lexers, ASTs, IR generation) to production AI pipelines (YOLOv8 inference, CNN training) to developer tooling (dependency-aware scaffolding, plugin architectures). The through-line is depth.

I'm a CS undergrad at Parul University who prefers shipping over theorizing. Technical Head at CDC. Published to npm. 10+ production systems deployed.

PRINCIPLE / 01

Depth over breadth

Understanding one layer deeply teaches you how every layer above it works. I invest in fundamentals because they compound.

PRINCIPLE / 02

Architecture is a product decision

How you structure code is how you structure future optionality. Good architecture makes teams faster, not slower.

PRINCIPLE / 03

Ship, then refine

Deployed imperfect systems teach more than perfect designs that never ship. Iteration is the mechanism, not a fallback.

PRINCIPLE / 04

Open source is how ideas scale

A CLI on npm reaches more developers than any closed system. Building in public creates accountability and compounds value.

§ 02 — Selected Work

Projects that
pushed limits.

PROJECT / 01 Active · npm published

Foundation CLI

A dependency-aware project scaffolding engine for the modern stack.

Built a full monorepo CLI toolchain published to npm as @systemlabs/foundation-cli. The core is a DAG-based PromptGraph engine that resolves module dependencies before writing a single file, an atomic FileTransaction system, and a plugin SDK that lets the community extend the engine via sandboxed lifecycle hooks. 28 built-in modules across 7 categories. 5 packages. Full TypeScript. CI/CD via GitHub Actions.

cli core / DAG modules (28) plugin-sdk FileTransaction (atomic) testing entry point PromptGraph
DAG-based conflict resolution before file writes
5 packages · 28 modules · 7 categories
@systemlabs/foundation-cli on npm
TypeScript Node.js pnpm monorepo Turborepo AJV validation GitHub Actions npm published
PROJECT / 02 Active · Systems Research

Dragoon

A JavaScript compiler written in C11. No shortcuts. No parser generators.

Built a full compilation pipeline from scratch in C11 — lexer, recursive-descent parser, AST construction, semantic analysis, intermediate representation, control flow graph, optimization passes (constant folding, dead code elimination), and QBE IR emission to native Linux binaries. This is the kind of project you build when you want to understand how languages actually work.

Lexer Parser recursive AST semantic IR / TAC CFG Optimize fold + DCE QBE IR backend Native Binary Linux/WSL · ELF JavaScript source executable
C11 — zero dependencies
8 stages from source to binary
Linux/WSL · QBE backend · ELF
C (C11) QBE Backend Systems Programming Compiler Theory CFG / DCE Linux/WSL
PROJECT / 03 Completed · Municipal Deployment

Smart Road Damage
Detection Platform

Full-stack AI system for smart-city pothole detection and civic reporting.

Built for municipal evaluation — a complete citizen-to-government pipeline. Citizen portal, admin console, YOLOv8 inference service in Flask, Node.js API, MongoDB backend, geolocation reporting, Firebase auth, Recharts analytics dashboards, and Leaflet map visualizations for government monitoring.

YOLOv8 inference via Flask microservice
MERN + Flask + Firebase + Leaflet
Municipal evaluation — real users
YOLOv8 Flask React Node.js MongoDB Firebase Leaflet
PROJECT / 04 Completed · 97.94% Accuracy

OCR Handwritten
Character Recognition

CNN trained on 884,900 samples with corrective feedback loop.

Trained a convolutional neural network from scratch to classify handwritten characters with 97.94% accuracy on 884,900 samples. Served inference via a Flask UI that supports freehand drawing input — and a corrective feedback mechanism to improve predictions over time.

884,900 samples
97.94% classification accuracy
Drawing input + feedback loop
TensorFlow / Keras CNN Flask Deep Learning OpenCV
§ 03 — Capabilities

The stack behind
the systems.

Domain / 01

Systems Engineering

C (C11) Compilers Lexer / Parser AST / IR CFG Analysis QBE Backend CLI Design Monorepo Architecture
Domain / 02

Full Stack & Backend

Node.js Express React Next.js Flask Django MongoDB PostgreSQL SQLite Firebase REST APIs WebSockets
Domain / 03

AI / ML

TensorFlow Keras YOLOv8 OpenCV CNN Design Model Training Inference Serving Prompt Engineering
Domain / 04

Dev Tooling & Infra

TypeScript pnpm Turborepo Docker GitHub Actions CI/CD Pipelines npm Publishing Plugin Systems AJV Validation
§ 04 — Open Source

Built in public.

@systemlabs/foundation-cli
Dependency-aware scaffolding engine with DAG-based PromptGraph, atomic file transactions, and a community plugin SDK. Published to npm.
TypeScript 5 packages · 28 modules
Dragoon — JS Compiler
A JavaScript compiler written in C11 from scratch. Full pipeline: lexer → AST → IR → CFG → QBE → native binary. No external dependencies.
C 8-stage pipeline
VMC Road Damage Detection
Smart-city pothole detection platform with YOLOv8 inference, citizen portal, admin console, and government analytics dashboard.
Python + JS Municipal deployment
CNN Letter Identification
Handwritten character recognition CNN trained on 884,900 samples. 97.94% accuracy with Flask inference UI and corrective feedback system.
Python 97.94% accuracy
Surakshita — Women Safety Analytics
Real-time threat detection platform using AI/ML + IoT with emergency alerts, GPS tracking, gesture-based triggers, and community reporting.
Python + IoT Real-time alerts
Multilingual Real-Time Chat
WebSocket-based chat platform with live message translation, multi-user support, hashed authentication, and SQLite persistence.
Python Flask-SocketIO

Let's build
something together.

I'm currently open to full-time roles, collaborations, and interesting open-source projects. If you're building something that requires genuine depth — reach out.

B.Tech CSE — Parul University
⬡ Technical Head, CDC
⬡ npm: @systemlabs/foundation-cli
ronak-parmar.vercel.app