How kaihag250/riscom works
Riscom extracts risk information from selected text, computes or evaluates risk measures, and presents communication guidance in a browser plugin. The documented primary workflow is visible in the plugin and both server variants: text is submitted for LLM inference, passed to a logic pipeline, and returned as advice. The repository also contains model fine-tuning, synthetic-data tools, evaluation, and experimental logic work. Server and logic paths are kept deployment-specific; unsampled subsystems are represented only at documented directory paths.
Main entry points: popup.js Plugin UI · server_bwCloud/server.py BWCloud API · server_runpod/server.py RunPod API · inference_worker.py BWCloud inference
Components 18
Browser plugin
- Plugin UI
plugin/popup.js - Selection handler
plugin/content.js
Inference runtime
- BWCloud API
server/server_bwCloud/server.py - BWCloud inference
server/server_bwCloud/inference_worker.py - BWCloud evaluator
server/server_bwCloud/logik_15_07.py - RunPod API
server/server_runpod/server.py - RunPod inference
server/server_runpod/inference_worker_1507.py - RunPod evaluator
server/server_runpod/logik_16_07.py
Model development
- Fine-tuned adapter
LLM/adapter - Fine-tuning
LLM/finetuning - Data generation tools
LLM/tools - Model evaluation
LLM/evaluation
Research logic
- Experimental evaluator
research/Logik_Formatierung/logic.py - Risk data schema
research/Logik_Formatierung/pydantic_model.py - Output deserializer
research/Logik_Formatierung/deserializer.py
Other components
- Reader
- Web page
- Mistral base model
Connections 25
- Reader → Web page: reads
- Web page → Plugin UI: selects text
- Plugin UI → BWCloud API: submits text
- Plugin UI → RunPod API: submits text
- BWCloud API → BWCloud inference: runs inference
- BWCloud API → BWCloud evaluator: evaluates output
- BWCloud inference → Fine-tuned adapter: loads model
- BWCloud inference → Mistral base model: loads base
- BWCloud inference → BWCloud API: returns fields
- BWCloud evaluator → BWCloud API: returns advice
- RunPod API → RunPod inference: runs inference
- RunPod API → RunPod evaluator: evaluates output
- RunPod inference → Fine-tuned adapter: loads adapter
- RunPod inference → Mistral base model: loads base
- RunPod inference → RunPod API: returns fields
- RunPod evaluator → RunPod API: returns advice
- BWCloud API → Plugin UI: returns advice
- RunPod API → Plugin UI: returns advice
- Plugin UI → Reader: displays guidance
- Selection handler → Web page: reads selection
- Fine-tuning → Fine-tuned adapter: trains adapter
- Data generation tools → Fine-tuning: supports data
- Model evaluation → Fine-tuned adapter: evaluates model
- Experimental evaluator → Output deserializer: loads risk data
- Experimental evaluator → Risk data schema: uses schema