Infrastructure for research continuity
This project connects documents, source code, numerical results and working notes into a personal research lifecycle archive. It is a response to work that spans simulations, electronics, local software and AI-assisted development.
The focus is continuity: preserving what was run, what changed, what can be reproduced and what still needs verification.
From a question to an archived result
- Inputs
Keep the starting point
Preserve source files, assumptions and a named baseline.
- Implementation
Keep the method
Version scripts, models, firmware and configuration with Git.
- Execution
Keep the evidence
Retain logs, machine outputs and the actual environment information.
- Analysis
Keep the reasoning
Connect figures and reports to their source data and analysis scripts.
- Handoff
Keep the next step clear
Record the current revision, limitations and remaining validation tasks.
Tools with defined roles
| Role | Current tools | Purpose |
|---|---|---|
| History & code | Git / GitHub | Version project changes and keep source work reviewable. |
| Development | VS Code · Codex | Editing, code assistance and engineering workflows. |
| Systems access | SSH · Linux · Raspberry Pi | Connect local development to embedded and system work. |
| Retrieval & local workspace | AnythingLLM / ResearchLLM | Explore document ingestion, search and a research-focused knowledge environment. |
| Model environment | Local + cloud LLM | Use different capabilities while keeping source verification explicit. |
A tool being connected does not establish that every workflow using it is reliable. The archive records actual behavior and its verification boundary.
ResearchLLM · ongoing development
The ResearchLLM direction combines a local research document environment with local and cloud language models. Development includes startup and shutdown scripts, document synchronization and the surrounding workflow.
Recent records include verification of document synchronization for add, modify and delete operations, alongside further startup-script repair. This does not establish mature scientific question answering, complete unattended operation, or numerical correctness of generated answers.
AI-assisted engineering with an evidence boundary
AI can help write code, organize records and inspect a project. The useful result still needs a traceable source and the right verification: static review, actual execution, numerical interpretation, or a hardware test.
- Keep generated artifacts and the baseline that produced them.
- Distinguish a model’s explanation from a measured or executed result.
- Use the actual current files to settle conflicting versions.
- Leave a compact project state so the next session can continue from evidence.
What makes a record reusable?
A reusable record explains its inputs, software environment, execution order, outputs and validation. A numerical figure should be tied to a script and dataset; an embedded release should identify the board and test conditions.
Automation is added where it removes repeat work, with checks for what it actually changes. It remains a developing part of the infrastructure rather than a promise that all projects are already automated.
Current records
Versioned projects & research workflow
Git-based archives, development logs, ResearchLLM integration and a growing research lifecycle repository.
The profile Gateway opens the verified user-specified GitHub account. No specific repository or release URL is fabricated.
Next practical steps
Make project state, baseline identity and validation records consistent across repositories. Connect public releases and documents when available, and expand retrieval only after its behavior is tested against real research questions.
