About Autoresearch
Alexandru DAN
CEO & Founder, TVL Tech
Creator of the Autoresearch system. Building autonomous AI-powered research tools that produce verified, citation-backed knowledge with zero human intervention.
Where AI meets your needs.
TVL Tech builds cutting-edge software solutions — AI, Web3 & Custom Software. Real solutions. No compromises.
- Artificial Intelligence — Transform business processes and customer experiences through AI-driven innovation
- Blockchain Infrastructure — Full-cycle development and deployment from concept to mainnet
- Custom Web Development — Solutions focused on seamless user experiences and optimized performance
Trusted by 12+ organizations across blockchain, AI, logistics, and enterprise. Autoresearch is TVL Tech's open-source proof that autonomous AI research is possible, auditable, and trustworthy.
What is Autoresearch?
Autoresearch is a single-subject autonomous research system. You define a research goal, and the system does everything else: generates research questions, searches the web, writes findings with inline citations, and verifies everything through a three-judge review panel.
The entire research process is transparent and auditable. Every finding has numbered citations [N] linking to verified sources. Three independent Claude Opus judges review every piece of content before it's published.
How It Works
↓
explode-goal.sh generates 30-50 research tasks
↓
research.sh picks tasks, runs Claude Code (web search + write)
↓
3 judges review: Evidence, Consistency, Completeness
↓
PR merged → findings published → new tasks proposed
↓
Score climbs toward 100%
The Score
The research score (0-100%) is computed from three components:
- Task completion (60%) — percentage of research Issues completed
- Citation quality (30%) — percentage of content paragraphs with inline [N] citations
- Confidence (10%) — penalizes LOW confidence sections
The Three Judges
| Judge | Focus | Checks |
|---|---|---|
| Judge 1 | Evidence | Fabricated facts, phantom citations, unsourced claims |
| Judge 2 | Consistency | Contradictions, date mismatches, style, confidence levels |
| Judge 3 | Completeness | Coverage gaps, missing perspectives, skeptical outsider view |
Technology
- Research engine: Claude Opus via Claude Code CLI (
claude -p) - Orchestration:
research.sh— bash loop managing branches, PRs, judges - Task management: GitHub Issues with labels for type, priority, status
- Backend: NestJS with hourly cron scheduler
- Quality: ShellCheck + automated tests (bash + Jest)
Source Code
Autoresearch is open source. The full code, research history, and all Issues are public.