nikitakorotaevnikita-sudo/class-og-final

How nikitakorotaevnikita-sudo/class-og-final works

This is a Russian citizen-appeals classification agent for the 59-FZ framework and official classifier, with Directum RX integration and operator review. Its main workflow accepts appeal text or an RX document, splits it into questions, retrieves classifier candidates, routes and classifies them, validates the result, then presents it for operator verification. The graph also captures configuration, audit feedback, and model-improvement capabilities. README details the classification stages; some subsystem wiring is inferred only where source excerpts are unavailable, so uncertain links are omitted.

Main entry points: api_server.py Classification API · classifier_agent.py Candidate retrieval · rx_client.py RX document client · section_router.py Section routing

Components 23

Intake and API

Classification API src/api_server.py
RX document client src/rx_client.py
Text extraction src/text_extractor.py

Classification

Question segmentation src/classifier_agent.py
Section routing src/section_router.py
Candidate retrieval src/classifier_agent.py
Classifier vector index src/build_vectordb.py
Optional rerankers src/reranker.py
LLM classification src/classifier_agent.py
LLM provider client src/integration_api.py
Result validation src/classifier_agent.py

Operator operations

Operator review UI src/static/app.js
Verification API src/api_server.py
Appeal audit log src/appeals_logger.py
Background job queue src/job_queue.py
Runtime settings src/settings_store.py

Data and improvement

Classifier annotations src/annotations_storage.py
Model fine-tuning src/auto_finetune.py
Historical import src/auto_import_historical.py

Other components

Appeal submitter
Review operator
Directum RX
LLM provider

Connections 25

Appeal submitter → Classification API: submits appeal
Review operator → Classification API: requests RX document
Classification API → RX document client: fetches document
RX document client → Directum RX: reads document
RX document client → Text extraction: extracts text
Classification API → Question segmentation: classifies text
Question segmentation → Section routing: routes topics
Question segmentation → Candidate retrieval: retrieves per question
Section routing → Candidate retrieval: filters candidates
Candidate retrieval → Classifier vector index: searches vectors
Candidate retrieval → Optional rerankers: optionally reranks
Candidate retrieval → LLM classification: supplies candidates
LLM classification → LLM provider client: requests classification
LLM provider client → LLM provider: calls model
LLM classification → Result validation: returns choices
Result validation → Classification API: returns validated result
Classification API → Operator review UI: serves result
Operator review UI → Review operator: presents alternatives
Review operator → Verification API: confirms or corrects
Verification API → Appeal audit log: records verification
Verification API → Classifier annotations: saves annotation
Appeal audit log → Model fine-tuning: supplies verified data
Classification API → Background job queue: dispatches jobs
Classification API → Runtime settings: reads configuration
Classification API → Historical import: starts historical import

More Python diagrams: simpleai-team/simpleai · anonnerdfrenzy/survival-llm · hrithikkoduri/webrover · sparklost/endcord · ajndkr/lanarky

Drawn by GitDiagram on September 29, 2026 from the repository's file tree, README and source files. Read as Markdown ·