Skip to content
View RANJITHROSAN17's full-sized avatar
🙃
🚀💰🐹🇮🇳
🙃
🚀💰🐹🇮🇳

Organizations

@Quasynvo

Block or report RANJITHROSAN17

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
RANJITHROSAN17/README.md

Ranjith Rosan — Analyst at Accenture, Machine Learning and Computer Vision, open-source contributor, Chennai, India

94 external pull requests 22 PRs merged upstream 8 PRs in review 52 external projects

LinkedIn GitHub followers Location Chennai, India

🌍 Open Source  ·  🚀 Projects  ·  💼 Experience  ·  📊 Stats  ·  📫 Contact


I build small, well-documented systems end to end — data, model, API, UI, deployment — and contribute fixes to the open-source tools I use. Analyst at Accenture; previously a computer vision intern and open-source contributor. I care about honest evaluation, reproducible fixes and maintainable tooling.

🔭 What I'm focused on now

  • MCP security & AI infrastructure — Vexar-MCP (static analysis, SARIF, policy gates for CI) and vexar-autofix
  • Agentic payments — Grace: judging an agent's spend against the business's own PayPal payment history instead of a static ceiling
  • Machine learning & computer vision — object detection, auto-labelling, model cards and honest evaluation
  • Research & systems code — maintenance fixes in large, unfamiliar codebases (ML research, shader transpilers, developer tools)
  • Building in public — every project below ships with docs, tests or a live demo, and states its limits clearly

🌍 Open-source contributions — by company / organisation

94 pull requests to 52 external projects · 22 merged upstream · 8 in review right now · auto-updated 06 Nov 2026 - every number below links to a live search you can verify.

Status key: 🟣 Merged = merged upstream  ·  🟢 In review = open PR, not yet merged - I don't claim orgs I haven't landed in (yet).

🟣 Merged upstream in: CrossGL · Ultralytics · Amazon Web Services · Opencollab MCP · Golang Cafe · +6 earlier in smaller projects

🟢 Currently in review at: Kubeflow

Organisation Project What my PRs touch 🟣 Merged 🟢 In review Latest PR
CrossGL logo CrossGL CrossGL/crosstl (★1.8k) Shader transpiler (GLSL/HLSL/Metal/SPIR-V/Slang/WGSL) - backend lowering, resource binding & provenance, conformance harness 10 · 🟣 #3050
kubeflow logo Kubeflow kubeflow/mcp-server (★58) MCP server for Kubeflow Trainer - rate-limit/circuit-breaker & RED metrics, schedulable-GPU counting, training lifecycle fixes · 8 🟢 #471
ultralytics logo Ultralytics ultralytics/yolov3 (★11k) YOLOv3 object detection - requirements fix 1 · 🟣 #2333
aws-samples logo Amazon Web Services aws-samples/step-functions-workflows-collection (★281) AWS Step Functions workflow collection - template update 1 · 🟣 #283
prakhar1605 logo Opencollab MCP prakhar1605/Opencollab-mcp (★43) Contribution-matchmaking MCP - shared GitHub paginator, closing-PR vs. plain-mention detection 2 · 🟣 #216
golang-cafe logo Golang Cafe golang-cafe/job-board (★228) Go job board - golangci-lint workflow updates 2 · 🟣 #121

Plus 6 earlier merged PRs in smaller projects - see every upstream-merged PR.

📂 Expand — every featured PR, grouped by organisation (auto-updated)

CrossGL logo CrossGL — CrossGL/crosstl · 10 merged · 0 in review

kubeflow logo Kubeflow — kubeflow/mcp-server · 0 merged · 8 in review

ultralytics logo Ultralytics — ultralytics/yolov3 · 1 merged · 0 in review

aws-samples logo Amazon Web Services — aws-samples/step-functions-workflows-collection · 1 merged · 0 in review

prakhar1605 logo Opencollab MCP — prakhar1605/Opencollab-mcp · 2 merged · 0 in review

golang-cafe logo Golang Cafe — golang-cafe/job-board · 2 merged · 0 in review

⚡ Latest merged & in-review pull requests


Find a project by what you want to do

Your goal Start here Evaluation boundary
Check an MCP server before shipping Vexar-MCP Static findings and coverage, not a security certification
Explore reviewable model-generated fixes vexar-autofix Start with mock mode and a disposable clone
Run an assistant on your own device Quasnooze Local model default; powerful tools need careful review
Try a lightweight Windows editor Morrow Unsigned early preview; use non-critical files
Learn everyday Kannada Kannada Saathi Browser learning aid, not a professional interpreter
Practice interview problem-solving DSA Mastery Plan Browser progress; export a backup

For research and notebook projects, read their data and evaluation limitations before interpreting metrics. Contribution status below distinguishes merged work from PRs still in review; an open PR is not an endorsement by its upstream organisation.

🚀 Featured projects

CardioLens 3D repository card Quasnooze repository card Vexar-MCP repository card vexar-autofix repository card Grace repository card Kannada Saathi repository card DSA Mastery Plan 2026 repository card Grounding DINO Auto-Labelling repository card YOLOv3 Custom Training repository card

Project What it is Stack ★ Language Updated
CardioLens 3D · live demo Research-only coronary-artery visualisation - interactive three.js schematic + scikit-learn baseline served from FastAPI, with a model card and documented limits (not for clinical use) React - three.js - FastAPI - scikit-learn 3 JavaScript 2026-11-02
Quasnooze A private, local-first personal AI assistant you message from Telegram, Discord, Slack, WhatsApp or the phone browser - 21 tools behind an approval gate, reminders that fire, token streaming, and a code-level honesty check Node.js - Ollama - MCP - PWA - zero runtime deps 13 JavaScript 2026-11-05
Vexar-MCP · live demo Security, schema and token-cost scanner for MCP servers - static analysis that never runs your code, CI policy gates, SARIF output, and an explicit 'n/a' when a scan could not read a file Python - MCP - SARIF - GitHub Actions 3 Python 2026-11-03
vexar-autofix · live demo An MCP security agent that has to prove its own patches: the finding must drop, nothing else may break, the target's pytest suite must still pass (Nebius x NVIDIA hackathon build) Python - NVIDIA Nemotron - Nebius Token Factory - MCP 3 Python 2026-11-02
Grace Agentic-payments judge - scores an agent's payment against the business's own PayPal payment history and cites the payments it reasoned from (PayPal AI Hackathon, sandbox only) Python - PayPal Orders API - payment graph 3 Python 2026-11-02
Kannada Saathi · live demo Offline-first, installable Kannada companion for everyday situations - 111 reviewed phrases with audio, strict TypeScript, 133 tests, 0 audit vulnerabilities (Kannada Digital Innovation Hackathon 2026) React - TypeScript - Vite - PWA 4 TypeScript 2026-11-02
DSA Mastery Plan 2026 · live demo 150-day, pattern-first DSA roadmap - 466 verified free LeetCode problems, 33 pattern playbooks, browser-only progress tracker vanilla JS - GitHub Pages 4 JavaScript 2026-11-02
Grounding DINO Auto-Labelling Zero-shot labelling for object-detection datasets - describe classes in English, review detections, export Pascal VOC PyTorch - Grounding DINO - supervision 4 Python 2026-11-02
YOLOv3 Custom Training Step-by-step guide + notebook for training YOLOv3 on your own dataset with Darknet in Google Colab Darknet - YOLOv3 - Colab 6 Jupyter Notebook 2026-11-02

Also public on this profile, just not featured above: linear-regression-machine-learning (Jupyter Notebook · ★5) · rock-paper-scissors (HTML · ★5) · musicgenre (Jupyter Notebook · ★4) · morrow-editor (TypeScript · ★3) · career-forge (Python · ★3) · LiveBridge (JavaScript · ★2)


💼 Experience

Role Organisation Focus
Analyst Accenture Machine learning & computer vision
Computer Vision Intern SeeWise.AI Industrial-safety detection — PPE, people and machinery detection for workplace safety
Open-source contributor Independent — upstream PRs across many projects (receipts in Open Source above) Bug fixes, metrics/observability, shader backend lowering, testing, docs

🟣 Merged upstream: CrossGL · Ultralytics · Amazon Web Services · Opencollab MCP · Golang Cafe · +6 earlier in smaller projects  ·  🟢 In review: Kubeflow  →  full breakdown


🤝 How I contribute

  • Reproduce first, patch second — every fix starts from the failing case, not a guess
  • Smallest defensible change — focused PRs a maintainer can actually review, with tests/conformance runs where the project has a harness
  • Review-friendly — I stay on the PR thread until it's merged or closed; the fastest way to reach me is right there

🛠️ Tools I work with

Core: Python (pandas, NumPy, scikit-learn, PyTorch, TensorFlow/Keras, OpenCV, FastAPI) · JavaScript/TypeScript (React, three.js, Vite) · C++ · SQL

MLOps & infra: Docker · Git & GitHub Actions · MCP (Model Context Protocol) · Kubeflow · Jupyter / Google Colab · Linux

AI security & agents: static analysis (AST-based scanners, SARIF, policy-as-code CI gates) · LLM agents on NVIDIA Nemotron via Nebius Token Factory · PayPal Orders API sandbox

Via open source I also read and patch code in Go, Java, Rust and shader languages (GLSL / HLSL / Slang / WGSL) when the project calls for it.

Python, PyTorch, TensorFlow, scikit-learn, OpenCV, FastAPI, React, TypeScript, JavaScript, Three.js, Vite, C++, Docker, Linux, Git, GitHub


📊 GitHub stats

GitHub stats for RANJITHROSAN17 Most used languages (my repos only, forks excluded)

Contribution streak

Top languages counts only my own repositories — the updater automatically excludes my forks of large upstreams (VS Code, Ghidra, Google Research, …) so the chart can't be skewed by other people's code.


About Ranjith Rosan

Ranjith Rosan maintains the original projects linked on this profile and contributes to other open-source projects. Each repository documents its purpose, setup and limitations; upstream contribution status is shown separately from project ownership.

📫 Contact

LinkedIn · open an issue on any repository above — I reply. If you're a maintainer reviewing one of my PRs: thank you, and the fastest way to reach me is right there.

Curious how this page maintains itself? docs/AUTOMATION.md explains the 2-hourly updater, its honesty rules, and how to reuse it.

🤖 This profile is auto-updated every 2 hours by GitHub Actions - last run 06 Nov 2026. On GitHub since 2021 · 17 public repos · 11 followers.

Pinned Loading

  1. Quasynvo/Quasnooze Quasynvo/Quasnooze Public

    Local-first personal AI voice assistant for web, Windows, Telegram, Discord, Slack & WhatsApp — Ollama + MCP tools, Fluid Motion audio visuals, permission-first. Node.js · zero runtime dependencies…

    JavaScript 13 5

  2. Vexar-MCP Vexar-MCP Public

    Security, schema and token-cost checks for MCP servers and AI agent tools. Static analysis that never runs your code, with policy gates for CI, SARIF, and a deliberate 'n/a' when a scan could not r…

    Python 3 2

  3. vexar-autofix vexar-autofix Public

    Reviewable MCP security patch proposals, checked by a pinned scanner, patch audits and target tests. Offline mock rehearsal or optional Nebius/Nemotron inference.

    Python 3

  4. Kannada-Digital Kannada-Digital Public

    Kannada Saathi: an offline-first Kannada phrase-learning PWA with bundled audio and everyday situations. Built for the Kannada Digital Innovation Hackathon 2026.

    TypeScript 4

  5. warden warden Public

    Grace: a sandbox agent-payment approval prototype using synthetic payment histories, explainable verdicts and optional PayPal integration. Not validated fraud detection.

    Python 3

  6. cardiolens-3d-risk cardiolens-3d-risk Public

    Research-only 3D coronary visualisation: interactive three.js schematic + scikit-learn baseline served from FastAPI, with a model card and documented limits. Not for clinical use.

    JavaScript 3