Hi 👋 I’m Rudransh. I build AI on my own GPU.
Backend & AI-systems engineer, currently a BCA student. I build local-AI agents and scalable retail ERPs that run directly on my own GPU. The 2am late nights are spent debugging deep system architecture, not writing shaders. Every repo I ship is built through curiosity, rigorous experimentation, and an unreasonable number of debugging sessions.
The Rig rudransh-ubuntu
- CPU
- Intel Core Ultra 7 265K · 3.9 GHz
- GPU
- NVIDIA RTX 5070 Ti · 16 GB
- RAM
- 32 GB
- Storage
- 2 TB NVMe · dual-boot
- OS
- Ubuntu 24.04 LTS · x86_64
The Laptop daily driver
- CPU
- Intel i5-12500H · 12C / 16T · 4.5 GHz
- GPU
- Intel Iris Xe (integrated)
- RAM
- 16 GB
- Storage
- 512 GB NVMe · Micron 2450
- OS
- Ubuntu 24.04 LTS · x86_64
Not a flex — a setup. Just where the scholarships, the late nights, and the savings ended up. The home office is the office now.
Who’s typing this.
I specialize in local AI architectures — building autonomous agents and scalable ERPs that run locally, avoiding the need to depend entirely on external APIs. My focus is on the backend engineering that makes this possible: robust schedulers, efficient message queues, and highly optimized data pipelines.
A recent challenge in AgentForge: agent handoffs could blow the 16 GB VRAM ceiling when models loaded concurrently. I serialize allocation behind a global async lock and a FIFO scheduler that keeps small models warm and evicts the heavy ones on demand — no cold-load tax, no race conditions. Solving deep systems problems is where I thrive.
The stack I rely on daily: Python, TypeScript, FastAPI, Next.js, PostgreSQL, and Ollama. My approach focuses on rigorous testing, systematic architecture, and pushing the limits of what consumer hardware can achieve.
Projects that kept me awake.
AgentForge
A VRAM-aware multi-agent code IDE built on Ollama. Seven local models, one consumer GPU, zero cloud.
Read case study
BusinessHub ERP
Retail ERP with 100K+ rows, dual-layer inventory, Z-Score anomaly detection, and a grounded Llama 3.1 assistant.
Read case study
PharmaCare
Healthcare e-commerce on Next.js 16 + React 19. Storefront, cart, checkout, admin dashboard — end-to-end.
Read case study
Brave Helper
Local-first AI side-panel for Brave / Chromium. Auto-router, AES-encrypted vault, RAG, vision — all on the user's own GPU via Ollama.
Read case studyNeural net visualizers, MCP chat applications, credit-risk models, and data pipelines — a collection of prototypes driving my technical growth.
All reposWhat’s on my keyboard.
Languages
- Python — daily
- TypeScript
- JavaScript
- SQL
- Java — academic
Backend & data
- FastAPI
- PostgreSQL
- SQLAlchemy
- Pydantic
- WebSocket
Frontend
- Next.js
- React 19
- Tailwind v4
- Zustand
- ReactFlow · Recharts
AI & the fun stuff
- Ollama
- Llama 3.1
- Multi-agent systems
- Three.js — visualization
- GSAP
A /now page — updated when life changes, not when I remember.
Inspired by Derek Sivers’ now pages. Honest snapshot. No marketing.