I build privacy-first, hardware-aware software—bridging low-level PC and mobile device internals with on-device AI, hybrid retrieval pipelines, and fault-tolerant Python tooling. From bare-metal diagnostics and bootloader workflows to quantized mobile LLMs, my focus is simple: run compute locally where practical, design for real-world network and filesystem failures, and ship utilities that hold up under everyday use.
- 📱 On-Device & Edge AI: Offline Android apps built with Kotlin and Jetpack Compose, leveraging LiteRT, Vulkan GPU delegates, and real-time camera OCR.
- 🔧 Firmware, Mobile & Systems Labs: Custom Android ROM and kernel deployments, Linux mobile OS testing (PostmarketOS / Ubuntu Touch / Termux), and DNS-level network filtering (Pi-hole / AdGuard).
- 🛡️ Applied AI & Hybrid Retrieval: Search pipelines pairing lexical (BM25) and dense embeddings with prompt-injection defenses and structured evaluation.
- ⚙️ Fault-Tolerant Automation: Zero-dependency provisioning scripts, concurrent media pipelines, and reversible filesystem tools with dry-run previews and rollback manifests.
🌟 Flagship Project:
Local-LLM-AI
An offline-first Android inference and camera OCR engine built to benchmark quantized LLM execution, memory pressure, and Vulkan GPU acceleration on real mobile hardware—with zero cloud dependence.
| System | Architectural Focus | Core Stack |
|---|---|---|
Local-LLM-AI |
Offline Android LLM execution, hardware memory profiling, and camera OCR workflows | Kotlin · Compose · LiteRT · Vulkan |
APEX-ATS |
Hybrid resume retrieval engine combining lexical + semantic search with input-sanitization safeguards | Python · BM25 · Embeddings · React |
Hyperos-Unlock-Script |
Zero-dependency automation suite for Xiaomi HyperOS bootloader unlock requests and device provisioning | Python · CLI · Device Provisioning |
PyStream-Downloader |
Resilient, concurrent stream capture engineered with exponential backoff and fragment recovery | Python · Async I/O · HLS/DASH |
TubeNotes |
Automated media-to-document pipeline turning educational videos into searchable, vector-crisp PDFs | Python · NLP · PDF Generation |
🧰 View Developer & Media Utilities (5 Repositories)
| Utility | Capability |
|---|---|
Image-to-PDF-Converter |
Desktop and web-deployed Python digitization and compression tool that cuts manual document processing time by ~80%. |
PyFile-Organizer |
Non-destructive directory automation with dry-run previews, operation logs, and JSON rollback manifests. |
Py-Image-Compressor |
Batch image compression and format conversion with perceptual quality controls. |
Anime-Recommender |
Gemini-powered content matching and structured recommendation workflow. |
Image-2-Pdf-Demo |
Multi-image preprocessing and clean document compilation pipeline. |
- Languages & Frameworks:
Python·Kotlin·Jetpack Compose·C·PowerShell·React·Vite·HTML/CSS/JS - AI & Edge Runtime:
Google AI Edge (LiteRT)·Vulkan GPU Delegates·BM25 + Dense Embeddings·Gemini Workflows - Systems, Hardware & NetOps:
Linux (Ubuntu, Kali, Termux)·Windows 10/11·PC Hardware Diagnostics·Custom ROMs & Bootloaders·Pi-hole / AdGuard DNS - Certifications & Community: NIELIT CHMT ('O' Level — Batch Topper) · NIELIT 'O' Level (IT) · NASSCOM Junior Software Developer · Google Gemini Educator · Active in PyDelhi & GDG Noida




