-0284C7?style=for-the-badge&logo=apple&logoColor=white&labelColor=0D1117){kind=icon}
neural_core:
operator: Shubhransh Gupta
primary_focus: Senior iOS Developer (Primarily)
open_source: Active Open Contributor & On-Device AI Systems Explorer
mission_status: READY — Actively seeking high-impact technical challenges to counter & conquer
core_systems: Swift 6.0 · Concurrency (Actors/Tasks) · SwiftUI & UIKit · Scalable Architecture
ai_explorations: On-Device LLMs/SLMs · Metal Kernels · MLX-Swift · CoreML · INT4/AWQ Quantization
proven_impact: Upstream Merges in Apple & Google Toolchains · 50M+ Endpoints Scaled · 99.9% Reliability
neural_uplink: Bengaluru, India 🇮🇳 (Synchronous Global Remote)🌌 "I am primarily a Senior iOS Developer who treats mobile applications not as static UI canvases, but as real-time, high-concurrency distributed engines. Alongside engineering high-scale native iOS systems across consumer products, I am an eager open contributor collaborating with AI researchers on on-device LLMs, Metal GPU acceleration, and local-first intelligence. I am relentlessly looking for the next tough technical challenge to conquer."
I actively contribute to AI researchers, open research groups, and frontier projects pushing the boundaries of machine intelligence:
| Research Frontier | Direct Focus & Implementation | Breakthrough Milestones |
|---|---|---|
| Edge & On-Device LLM Inference | Quantization algorithms (INT4, AWQ, GGUF), KV-cache compression, memory alignment, and context optimization on mobile/desktop hardware. |
Sub-millisecond token execution, reduced memory footprints for 3B-7B parameter models. |
| Metal GPU & Silicon Acceleration | Low-level kernel optimization using Metal Shading Language (MSL) & mlx-swift for unified memory architectures. |
Merged patch resolving convolution padding bounds in Metal GPU kernels (mlx-swift#467). |
| Autonomous Agentic Runtimes | Multi-agent state orchestration, dynamic tool synthesis, and local-first context filtering for AI coding companions. | Built toknt (token optimizer for Cursor/Claude) & scavengers-assemble. |
| Privacy-First Local AI | Zero-telemetry vector indexing, in-browser neural document analysis, and client-side vector search. | Built findit (local browsing embeddings) & OfflineCV. |
⚡ tokntAI Context & Token Optimizer for Autonomous Agents High-efficiency, local-first context compression engine for frontier AI coding assistants (Cursor, Claude Code, OpenAI Codex). Dynamically parses AST trees and strips redundant context, cutting prompt latency and token burn by up to 60%.
|
Native macOS HTTP(S) Debugging & Network Proxy Next-generation proxy engine engineered in pure Swift 6.0. Features selective on-the-fly SSL decryption, zero-config device discovery across local subnets, real-time mock breakpoint rules, and full HAR stream exports.
|
|
Multi-Agent Frontier AI Simulation Engine Real-time 3D isometric simulation engine orchestrating multiple autonomous frontier models (Claude 3.5, Gemini 1.5/2.0, GPT-4o, Grok & local Ollama) interacting as cooperative autonomous agents within dynamic virtual worlds.
|
🔎 finditPrivacy-Preserving Local-First Semantic Search Personal semantic search engine for the internet you've already experienced. Embeds, indexes, and retrieves browsed knowledge 100% locally on your machine with zero server-side telemetry or cloud leakage.
|
|
Multimodal Anti-Gridlock Transit Engine Computational routing engine built to tackle severe urban bottlenecks. Pairs walking bypasses, suburban rail, and metro skybridges to bypass notorious traffic gridlocks with sub-second pathfinding.
|
60 FPS 3D Open-World Procedural Simulation Browser-based 3D procedural simulation featuring rigid-body vehicle physics, aerial dynamics, and real-time procedural street grid generation across 3,600+ lines of pure JavaScript without external 3D asset downloads.
|
Active open-source contributor solving core compiler, GPU runtime, and enterprise infrastructure challenges:
| Ecosystem / Organization | Target Repository | Engineering Breakthrough | PR / Commit Status |
|---|---|---|---|
| 🍎 Apple Silicon / AI | ml-explore/mlx-swift |
Fixed convolution padding bounds & memory bounds in Metal GPU kernels | MERGED #467 |
| 🍎 Apple Swift Toolchain | apple/swift-argument-parser |
Resolved core flag-parsing edge case for attached values (--opt=) in Swift CLI |
MERGED #961 |
| 🍎 Apple Protocols | apple/swift-protobuf |
Implemented useDeterministicOrdering in TextFormat serialization for reproducible testing |
MERGED #2159 |
| 🌐 Google Core Security | google/osv-scanner |
Fixed vulnerability report checklist filter state preservation in HTML output | MERGED #3049 |
| 📱 AppLovin / Cross-Platform | AppLovin/AppLovin-MAX-React-Native |
Resolved ad view constraints, sizing boundaries, and mount memory leak lifecycle | PR #488 #487 |
| 📱 AppLovin / Mobile | AppLovin/AppLovin-MAX-Flutter |
Updated Android compileSdk 36 compatibility across Flutter MAX plugin suite |
PR #366 |
| 🚀 Fastlane Mobile Toolchain | fastlane/fastlane |
Prevented duplicate prefix injection in application-identifier entitlements |
MERGED #30169 |
| 💻 Microsoft Cloud | dotnet/aspire |
Standardized telemetry timestamp controls across distributed cloud hosts | ACTIVE #19931 |
╔═══════════════════════════════════════════════════════════════════════════════════════════╗
║ SYNAPTIC TECH CAPABILITIES ║
╠═══════════════════════════════════════════════════════════════════════════════════════════╣
║ [Frontier AI & ML] MLX-Swift · Metal (MSL) · CoreML · PyTorch · INT4/AWQ/GGUF ║
║ Tokenizers · KV-Cache · ONNX Runtime · Autonomous Multi-Agents ║
║ [Systems & Languages] Swift 6.0 · Python 3.12+ · C / C++ · Rust (Exploring) · TypeScript ║
║ [iOS & Apple Core] SwiftUI · UIKit · Swift Concurrency (Actors, Async/Await) · SPM ║
║ [Distributed & Web] FastAPI · WebGL · Three.js · Cannon Physics · SSL/TLS Proxy Engines║
║ [DevOps & Tooling] GitHub Actions · Fastlane · LLDB · Instruments · Docker · CI/CD ║
╚═══════════════════════════════════════════════════════════════════════════════════════════╝
|
|
|
|
|
|
