How fareedkhan-dev/building-claude-from-scratch works
This repository is an executable notebook reproducing an agentic harness on an open-source model, with a research-paper replication as its end-to-end task. The documented workflow combines reasoning, orchestration, tool-grounded execution, reliability, stateful memory, and verification to produce a mechanically graded result. The graph follows README-level relationships only: no notebook source excerpts were provided, so internal calls and data transfers cannot be verified.
Main entry points: claude_from_scratch.ipynb End To End Run
Components 16
Cognition
- Cognitive Substrate
claude_from_scratch.ipynb - Frontier Patterns
claude_from_scratch.ipynb
Orchestration
- Reasoning Topology
claude_from_scratch.ipynb - Tool Execution
claude_from_scratch.ipynb - Stateful Orchestration
claude_from_scratch.ipynb - End To End Run
claude_from_scratch.ipynb - Specialist Subagents
claude_from_scratch.ipynb - Master Loop
claude_from_scratch.ipynb
Reliability
- Production Reliability
claude_from_scratch.ipynb - Budget Guard
claude_from_scratch.ipynb
Grounding And Trust
- Grounding And Evaluation
claude_from_scratch.ipynb - Pytest Verdict
claude_from_scratch.ipynb
Other components
- Research Task
- Dengue Paper
- DATASUS Data
- Open Model
Connections 15
- Research Task → End To End Run: runs task
README.md - Dengue Paper → End To End Run: replicates findings
README.md - DATASUS Data → End To End Run: grounds task
README.md - Open Model → End To End Run: powers harness
README.md - Cognitive Substrate → End To End Run: composes into
README.md - Reasoning Topology → End To End Run: composes into
README.md - Tool Execution → End To End Run: composes into
README.md - Production Reliability → End To End Run: composes into
README.md - Frontier Patterns → End To End Run: composes into
README.md - Stateful Orchestration → End To End Run: composes into
README.md - Grounding And Evaluation → End To End Run: composes into
README.md - End To End Run → Specialist Subagents: includes
README.md - End To End Run → Master Loop: orchestrates
README.md - End To End Run → Budget Guard: meters steps
README.md - End To End Run → Pytest Verdict: produces
README.md