salt-nlp/sparkme

How salt-nlp/sparkme works

SparkMe is a multi-agent system for semi-structured, multi-turn interviews. Its principal workflow accepts an interviewee’s text or optional speech input, runs an interview session in which an interviewer asks questions and planning and agenda components guide topic exploration and note-taking, then stores session state and logs. The README also documents a web interface with authentication and session management. The graph focuses on the primary src implementation, not the separate baselines or evaluation tooling. Some wiring is supported only by the SOURCE INDEX or README, so unsampled relationships are omitted or marked unverified.

Main entry points: interview_session.py Interview session · agenda_manager.py Agenda manager · session_agenda.py Session agenda · user.py User participant

Components 20

Interfaces and session

Terminal interface src/main.py
Web interface src/main_flask.py
Interview session src/interview_session/interview_session.py
Session messages src/interview_session/session_models.py
User participant src/interview_session/user/user.py

Interview agents

Interviewer src/agents/interviewer/interviewer.py
Agenda manager src/agents/agenda_manager/agenda_manager.py
Exploration planner src/agents/exploration_planner/exploration_planner.py
Agent framework src/agents/base_agent.py

Interview domain

Session agenda src/content/session_agenda/session_agenda.py
Topic manager src/content/session_agenda/interview_topic_manager.py
Question bank src/content/question_bank/question_bank_vector_db.py
Question records src/content/question_bank/question.py
Session memory src/content/memory_bank/memory_bank_vector_db.py
Memory records src/content/memory_bank/memory.py

Runtime services

Embedding service src/content/embeddings/embedding_service.py
Model engine src/utils/llm/engines.py
Speech input src/utils/speech/speech_to_text.py
Session logs src/utils/logger/session_logger.py

Other components

Interviewee

Connections 27

Interviewee → Terminal interface: starts interview README.md
Terminal interface → Interview session: starts session src/main.py
Interviewee → Web interface: uses web interview README.md
Web interface → Interview session: uses session src/main_flask.py
Interview session → User participant: creates participant src/interview_session/interview_session.py
User participant → Interview session: adds response src/interview_session/user/user.py
Interview session → Interviewer: sets up agent src/interview_session/interview_session.py
Interview session → Agenda manager: sets up agent src/interview_session/interview_session.py
Interview session → Exploration planner: sets up agent src/interview_session/interview_session.py
Interview session → Session agenda: loads agenda src/interview_session/interview_session.py
Interview session → Session memory: sets up memory src/interview_session/interview_session.py
Interview session → Question bank: sets up bank src/interview_session/interview_session.py
Interview session → Session messages: uses message model src/interview_session/interview_session.py
Interviewer → Agent framework: inherits from src/agents/interviewer/interviewer.py
Agenda manager → Agent framework: inherits from src/agents/agenda_manager/agenda_manager.py
Agent framework → Model engine: invokes engine src/agents/base_agent.py
Agenda manager → Session agenda: adds questions src/agents/agenda_manager/agenda_manager.py
Agenda manager → Memory records: uses memories src/agents/agenda_manager/agenda_manager.py
Session agenda → Topic manager: manages topics src/content/session_agenda/session_agenda.py
Session agenda → Question records: records questions src/content/session_agenda/session_agenda.py
Topic manager → Question records: uses questions src/content/session_agenda/interview_topic_manager.py
Question bank → Question records: stores questions src/content/question_bank/question_bank_vector_db.py
Session memory → Memory records: stores memories src/content/memory_bank/memory_bank_vector_db.py
Session memory → Embedding service: embeds for search src/content/memory_bank/memory_bank_vector_db.py
Question bank → Embedding service: embeds for search src/content/question_bank/question_bank_vector_db.py
User participant → Speech input: transcribes voice src/interview_session/user/user.py
Interview session → Session logs: logs session src/interview_session/interview_session.py

Drawn by GitDiagram on November 10, 2026 from the repository's file tree, README and source files. Read as Markdown ·