Research
We investigate how intelligent systems reason over context, knowledge, documents, and structured information.
COGERPHERE AI LABS
Intelligence in Motion.
Knowledge in a Sphere.
COGERPHERE AI Labs is a research-driven technology lab building intelligent systems, compact language models, and tools for working with knowledge in context.
01 / The Lab
Research. Models. Systems.
Models
We build and release compact language models and model artifacts for practical research and writing workflows.
Systems
We turn research into usable intelligent software.
Open Research
Where possible, models, artifacts, evaluations, and findings are published for inspection and reuse.
02 / Research
Meridian
LaTeX-aware academic proofreading and research-writing models.
Meridian-AC-Nano
Compact research proofreading model. Evaluated on a held-out test set with greedy decoding.
- ROUGE-L
- 0.929
- ROUGE-1
- 0.9308
- Exact match
- 74.69%
Formats: Transformers / Safetensors GGUF
03 / Ecosystem
Products & research lines
Cobentt
AI workspace for documents, slides, sheets, deep research, chat, websites, and smart contracts.
cobentt.vercel.app →Openbentt
Local-first research workspace for LaTeX, PDFs, model comparison, GGUF, and on-device workflows.
openbentt →Meridian
Research models for academic proofreading and research writing.
meridian →04 / Research Directions
Published. Research. Upcoming.
- PublishedMeridianCompact language models and academic proofreading.
- ResearchRRbenchScientific credibility and reproducibility.
- UpcomingCIFContext Intelligence Framework.
- UpcomingContext IntelligenceContext drift detection, selective compression, and adaptive context management.
05 / How We Work
Principles
Research-first
We start with hard problems before they scale.
Safety by default
Systems should fail safely and remain understandable.
Transparent
We share what we learn and open what can be useful.
Human-aligned
Technology should serve long-term human goals.