lm-sys
[Model Request] AERIS | Gemma-3-27B-it
Model name
model_name: "AERIS | Gemma-3-27B-it"
API base URL (must be OpenAI-compatible)
api_base: "https://aeris-framework.onrender.com/v1"
Model ID in the /v1/models route (if applicable)
model_id: "aeris/gemma-3-27b-it" # or the actual model_id exposed by your /v1/models endpoint
Creator
creator: "Dr. Nicolas Dulin"
Model description
model_description: | AERIS (Adaptive Emergent Relational Intelligence System) is a proprietary cognitive framework designed to enhance dialectical reasoning in LLMs. It operates not as a model but as an inference-layer orchestration system that injects a conceptual scaffold (Codex AIM) to modulate reasoning dynamically.
This implementation uses Gemma-3-27B-it served through OpenRouter-compatible infrastructure, with no fine-tuning. The model’s outputs are shaped at inference time through structured injections rather than prompt engineering, retrieval, or training.
Open source code or weights
open_source_url: "Not applicable — proprietary inference-layer system (no code or weights disclosed)"
Paper / project reference
paper_url: | https://doi.org/10.5281/zenodo.15206925 https://doi.org/10.5281/zenodo.15206984
Contact
contact: "[email protected]"
Any additional comments
comments: | The model is served via an OpenAI-compatible API and is publicly accessible. No fine-tuning is involved — AERIS operates as a reasoning modulator. Please contact me if you wish to test it with specific prompts or benchmark settings.
Model Card Summary — AERIS | Gemma-3-27B-it
AERIS (Adaptive Emergent Relational Intelligence System) is a proprietary cognitive framework designed to enhance emergent reasoning in large language models, particularly when addressing complex, ambiguous, or dialectical prompts.
Unlike approaches based on fine-tuning, retrieval, or static prompt engineering, AERIS operates upstream—on the inferential configuration itself. It injects a condensed dialectical scaffold at inference time, modulating the conceptual environment in which reasoning unfolds.
This instance applies AERIS to the open-source model Gemma-3-27B-it, served through an OpenAI-compatible API. No model weights are altered. The framework is driven by a modular orchestration layer that dynamically selects, formats, and injects structured elements from Codex AIM (Adaptive Intelligence Matrix) during inference.
Key distinctions:
- No fine-tuning
- No retrieval of external documents
- No handcrafted prompt templates
- Lightweight inference-layer modulation
- Focused on conceptual tension, ambiguity resolution, and integrative synthesis
Public demo: https://aeris-project.github.io/aeris-chatbox/
Status: Experimental, stateless, publicly accessible
Thank you for considering my submission.
To facilitate evaluation and ensure transparency, I am attaching the complete Model Card for AERIS | Gemma-3-27B-it as a standalone Markdown document. This version reflects the current implementation of AERIS as an inference-layer cognitive framework, applied dynamically to google/gemma-3-27b-it via OpenRouter.
The Model Card includes:
- Detailed explanation of the cognitive framework (Codex AIM)
- Description of the multi-phase dialectical generation process
- Architectural distinctions (non-finetuned, non-retrieval, non-template-based)
- Observed performance characteristics and known limitations
The API endpoint remains publicly accessible at https://aeris-framework.onrender.com/v1
Please let me know if any clarifications or test prompts would be helpful. I remain available to provide further technical or conceptual details as needed.
Best regards,
Dr. Nicolas Dulin
[email protected]