Code · architecture · study mode

Amardeep Singh Sidhu
About

Hi, I'm Amardeep Singh Sidhu

FireHacker.
Founder & AI Researcher @ AIEdx

I built FBA Lab so you can see how small and large language models are actually trained. Real code, visualization, and content, in the browser. I also ship FetchLens.ai and BubblSpace, and run the open First Break AI cohort.

What I'm building

Right now a lot of my focus is FetchLens.ai, making AI-driven traffic visible. ChatGPT, Claude, Perplexity, coding agents, and scrapers leave fingerprints most analytics never show: many bots never run your client-side tags, so dashboards look quiet while AI is already reading your pages. FetchLens closes that gap with a small install (script tag or Next.js middleware) so you can see which agents visit and what classic analytics misses.

I also build infrastructure for agents that remember, learn, and grow: agentic AI, skill portability, and multi-agent collaboration. DevOps is for code. MLOps is for models. SkillOps is for agents. BubblSpace mines skills from real work and deploys them to any agent runtime, so your agents do not start from scratch every time.

FBA Lab is the teaching side of that work. Walk a real training run through four synced views: code, visualization, content, and the run itself. The Speedrun track follows nanoGPT optimization from baseline through record jumps. Qwen C traces GGUF loading, buffers, the forward pass, tokenizer, and generation in plain C. Study mode lets you go at your own pace. No GPU required.

My journey

  • 2026My focus is sharply on core AI training. I am going deep into code, infrastructure, pipeline, and research.
  • 2025Deep research into multi-agent orchestration, reasoning models, and skill portability across AI toolchains.
  • 2024Launched BubblSpace with AI Personas, real-time voice, and enterprise workflow execution.
  • 2022Founded AIEdx, focusing on AI-powered solutions.

What I care about

  • AI traffic & observability. What analytics misses when agents and scrapers do not behave like browsers, and how to surface that without a bloated install.
  • SkillOps architecture. Mining skills from real work and deploying them across Cursor, Claude Code, Codex, and any SWE agent.
  • Multi-agent systems. Sub-agent orchestration where specialized agents work a problem from different angles. Every finding is cited.
  • Reasoning models. How agents move from pattern-matching to genuine strategic thinking.
  • Agentic workflows. Systems where AI does not just respond. It initiates, researches, and comes back with something useful.

Prompts fade. Skills compound.

I want agents that are persistent (they remember where you got stuck), social (they exchange skills with other Personas), portable (a skill built in one place travels), and compounding (the gap between a fresh agent and your agent should widen every day).

Let's connect

I like talking with builders, researchers, and people learning AI in public. Reach me on X, GitHub, or email contact@bubblspace.com.

AIEdxbubblspace.com

FBA Lab

This site is a four-mode visualization lab for real LLM training and inference. Code, viz, content, and runs. No GPU required.

FetchLens.ai

See AI agent traffic your analytics miss. Many agents never run your front-end JavaScript.

BubblSpace

SkillOps for AI agents. Personas that remember, pick up portable skills, and grow across runtimes.