name: "Istiaque Ahmmed"
alias: "The Dark Architect"
role: "AI Engineer & Interpretability Researcher"
location: "Toronto, Canada"
focus:
- Mechanistic Interpretability
- Transformer Internals & Feature Extraction
- Sparse Autoencoders (Anthropic-style)
- Multi-Agent Orchestration
- RAG Systems & Enterprise AI
currently_summoning:
- Transformer from Scratch (pure NumPy)
- Sparse Autoencoder (feature extraction)
- Circuit Finder (automated circuit discovery)
- Logit Lens & Tuned Lens
dark_philosophy:
- Understand AI from the inside out
- Build everything from scratch
- Production-grade code with blood tests |
I design and deploy production AI systems from the shadows:
Every project is built to be deployed, not demoed. 
|
| Project |
What It Does | Tech | Links |
|---|---|---|---|
| Enterprise AI Agent Platform | Multi-domain agent system with specialized agents for finance, legal, and healthcare. | Python, Cohere, FastAPI, Next.js | Repo · Live |
| AI Decision Firewall | Runtime governance layer that intercepts AI outputs before they become actions. | Python, FastAPI, Pydantic | Repo · Live |
| Prompt Injection Detector | Real-time detection of prompt injection attacks with pattern matching & heuristics. | Python, Regex | Repo · Live |
| Agent Sandbox | Secure code execution environment for AI agents with permission system. | Python, AST | Repo · Live |
| Model Diff | Compare model checkpoints to understand training changes with pattern detection. | Python, NumPy | Repo · Live |
| LLM Gateway | Multi-provider routing, fallback, rate limiting, caching for LLM APIs. | Python, asyncio | Repo · Live |
| MCP Tool Server | Production MCP server with multiple tools (file, shell, search). | Python, MCP | Repo · Live |
| Transformer Lab | GPT transformer from scratch in pure NumPy with interpretability probes. | Python, NumPy | Repo |
| Sparse Autoencoder | Extract interpretable features from neural network activations (Anthropic-style). | Python, NumPy | Repo |
| Circuit Finder | Automated circuit discovery for mechanistic interpretability. | Python, NumPy | Repo |
| Induction Head Detector | Find and analyze induction heads in any transformer. | Python, NumPy | Repo |
| Activation Steering | Control model behavior by modifying activations at runtime. | Python, NumPy | Repo |
**Open to Interpretability Research Engineer, Applied AI Engineer, and Forward Deployed Engineer roles.**
**Toronto, Canada | Open to relocation**
