Gopi Trinadh Maddikunta

Gopi Trinadh Maddikunta

Copyright @ 2026 GT Groups.
All rights are reserved.

From the Lab
What I Build
& Learn
Deep dives into AI, open-source journeys, and everything I learn building real systems.
⚑ Live Data
πŸ“Š Nova Bench Β· DGX Spark Benchmarks

How Fast Can a $4,699 Desktop Run LLMs?

11 models from 8B to 123B tested on NVIDIA DGX Spark. Interactive dashboard with real benchmark data.

⚑ 42.9 tok/s peak 🧬 11 models πŸ‘οΈ 90B vision πŸ”₯ 10K reasoning tokens
Explore Benchmarks β†’

Gemma 3 from Scratch

Screenshot 2026-02-24 104506


🧠 Gemma 3 from Scratch: I pre-trained a Gemma 3 architecture model from scratch and a 164M parameter language model built entirely from zero, trained on the TinyStories dataset to generate coherent children’s stories.

🌟 This journey has been a deep dive into transformer internals, from implementing Multi-Query Attention and RoPE embeddings to watching a model learn language one training step at a time…

Building BUD: From Zero to Production AI Agent



πŸš€ Building BUD: From Zero to Production AI Agent

I built BUD (Build, Understand, Deploy) β€” a self-hosted multi-platform AI assistant from scratch, serving 5 platforms simultaneously with 24+ MCP tools, a 5-layer memory system, and real-time voice I/O.

🌟 This journey has been a deep dive into AI agent architecture, multi-platform engineering, and production-grade deployment β€” all without LangChain or any agent framework…

Web Design

Firogo Ltd

Web Design

Firogo Ltd

Web Design

Firogo Ltd

My GSoC 2025 Journey

πŸš€ Contributing to Open-Source AI with Scala Center

I was selected as a Google Summer of Code 2025 Contributor with the Scala Center, where I worked on enhancing LLM4S β€” an open-source project focused on tracing and monitoring large language model workflows.

🌟This journey has been a deep dive into LLM tracing, open-source collaboration, and real-world AI development…