Zochi (Intology)¶
external · status: active · focus: end-to-end · discipline: computer-science · started: 2025
Project page: https://github.com/IntologyAI/Zochi
Source: projects/landscape/zochi.yml
Positioning¶
An end-to-end "artificial scientist" system from Intology, claimed to span hypothesis generation through to peer-reviewed publication. Differentiates itself from earlier AI-scientist releases by publishing the outputs — accepted ACL 2025 and ICLR 2025 workshop papers (CS-ReFT, Tempest/Siege) with reported state-of-the-art results — rather than only the pipeline.
Distinctive contribution¶
The strongest external-validation claim in the AI-scientist landscape: peer-reviewed acceptances of papers produced by the system, including ACL 2025 main proceedings. The repository releases code and a technical report covering an earlier version of the system.
Evaluation scores¶
| Dimension | Score (0–3) | Note |
|---|---|---|
| Lifecycle coverage | 3 | Seven stages from hypothesis through review; full lifecycle in claimed scope. |
| Autonomy level | 3 | Autonomous end-to-end discovery is the stated design target. |
| Architectural transparency | 2 | Public repository covers earlier version; current capabilities described in blog posts; full current pipeline not open. |
| Inputs supported | 2 | Research-area inputs with optional dataset; commercial back-end at intology.ai. |
| Outputs / reproducibility | 2 | Published papers + benchmark code; full system reproduction depends on current closed components. |
| Internal evaluation | 3 | External peer review at ACL 2025 + ICLR 2025 workshops — strongest external validation in the catalog. |
| Openness | 2 | MIT-licensed code for earlier version; current system features are commercial. |
| Maturity / traction | 2 | 305 stars; commercial backing (Intology); credible publication trajectory. |
| Cross-family policy | 0 | Same-model self-refinement is the canonical failure mode ARIS Table 4 identifies. |
| Runtime assurance | 2 | External peer-review acceptance (ACL 2025) is the strongest runtime-equivalent assurance signal in the catalog. |
| Cross-platform portability | 1 | Public code covers earlier version; current pipeline is commercial. |
Scored on 2026-05-18. See the evaluation rubric.
Tags¶
Pipeline stages: hypothesis-generation research-design data-analysis code-generation paper-drafting revision-editing referee-simulation
Architectural features: multi-agent tool-use iterative-loop artifact-versioning
Inputs: research-area
Outputs: paper-draft code experiment-results
Data sources: benchmark-datasets
Knowledge sources: literature
Limitations¶
- Public code lags the current capabilities described in marketing — current pipeline is not fully open.
- Validation is via specific peer-reviewed papers; cross-domain generality is asserted but not separately tested.
- Last push 2025-11; the open repository may not reflect the live system.
Related projects in this catalog¶
Also compared in¶
- A Survey of AI Scientists (
tie2025aiscientistsurvey) — Covered as an end-to-end system with peer-reviewed publication trajectory.
Related references (literature catalog)¶
- Wu, J. et al. (2025). Agentic Reasoning: A Streamlined Framework for Enhancing LLM Reasoning with Agentic Tools
wu2025agenticreasoning - Gartenberg, C. et al. (2026). More Versus Better: Artificial Intelligence, Incentives, and the Emerging Crisis in Peer Review
gartenberg2026morebetter