class Dhurjati:
location = "Panvel, India"
builds = "local-first AI products: the models run on my own GPUs and the data stays on the machine"
focus = ["backend engineering", "backend hardening", "stress & scrutiny testing", "systems thinking"]
ai_stack = ["Gemma 3", "llama.cpp", "vLLM", "Ollama", "pgvector RAG", "Claude API"]
home_lab = "two GPU PCs behind one OpenAI-compatible gateway with automatic failover"
works_on = ["Linux (Arch)", "macOS", "Windows", "Docker"]
def how_i_work(self):
return ("I use AI tools like ChatGPT and Claude to build faster. "
"The thinking is mine: the ideas, finding and solving the problems, "
"the backend hardening and the scrutiny testing.")|
Podcast Q&A Studio is an AI-powered tool designed to assist in podcast production. It helps users write, check, and time podcast questions, or provides mock interviews with coaching. |
AI cost estimates from machine drawings, inside a self-hosted workspace for startups: bookings with WhatsApp reminders, inventory alerts and a client portal. |
|
Test a startup's strategy in a simulated world: agent-based consumers, companies and traders on a 3D globe, calibrated to live data, with Monte Carlo forecasts and daily AI insights. |
School and tuition-centre management: live classes with recording, RAG-generated notes and tests, and an AI tutor that answers from paused video frames. |
|
Find vulnerable Python dependencies before they find you: real OSV.dev advisories, a local Gemma review with a Claude fallback, and every upgrade re-verified. |
A local visiting-card scanner: OpenCV straightens the photo, Apple Vision reads it, Gemma pulls out the contact, and pgvector powers typo-tolerant search. |
|
AI video intelligence: summaries from key frames plus Whisper captions, streamed live over Socket.IO, rate-limited and CSRF-protected, and stress-tested with Locust. |
One OpenAI-compatible link for two GPU PCs with automatic failover. It asks for the other PC's IP at boot, starts everything at logon and is reachable through a Cloudflare tunnel. (v1) |
|
My portfolio site: an AI reads every README, sorts each project into social impact or industry, writes the summary, and keeps it all up to date automatically. |
Languages
Backend, data & web
AI & machine learning
Platforms, DevOps & tools
All my projects are student-built and shared as-is, with no warranty or liability. Most are MIT-licensed; see each repo's LICENSE.