How sanelemanyela/academic-truth-engine works
Academic Truth Engine is a local, notebook-driven research assistant for answering questions from academic PDFs. Its principal workflow ingests and validates a PDF, recovers text with OCR when needed, creates citation-oriented semantic chunks, persists them in Chroma, retrieves evidence, and generates a grounded answer. Users can interact through a notebook Q&A loop or the injected widget UI; optional outputs include visual panels and structured briefs. The notebook’s implementation was not sampled, so those stage relationships are based on the README; source-level evidence is available for the UI query call and the local retriever packs.
Main entry points: Academic_Truth_Engine_v3.ipynb Research notebook · mempalace_db Chroma memory · ate_ui.py Q&A widget · source_material.pdf Academic PDF
Components 15
Research workflow
- Research notebook
Academic_Truth_Engine_v3.ipynb - Academic PDF
academic_data/source_material.pdf - PDF ingestion
Academic_Truth_Engine_v3.ipynb - Text and OCR
Academic_Truth_Engine_v3.ipynb - Semantic chunks
Academic_Truth_Engine_v3.ipynb
Memory and retrieval
- Chroma memory
mempalace_db - Query rewriting
query_rewriting_pack/llama_index/packs/fusion_retriever/query_rewrite/base.py - Hybrid fusion
query_rewriting_pack/llama_index/packs/fusion_retriever/hybrid_fusion/base.py - Evidence retrieval
Academic_Truth_Engine_v3.ipynb
Answer and outputs
- Grounded answer
Academic_Truth_Engine_v3.ipynb - Model inference
- Visual panels
Academic_Truth_Engine_v3.ipynb - Brief export
Academic_Truth_Engine_v3.ipynb
User interface
- Q&A widget
ate_ui.py
Other components
- Researcher
Connections 17
- Researcher → Q&A widget: asks questions
ate_ui.py - Researcher → Research notebook: runs workflow
ReadMe.md - Q&A widget → Research notebook: calls query
ate_ui.py - Academic PDF → PDF ingestion: provides PDF
ReadMe.md - PDF ingestion → Text and OCR: extracts text
ReadMe.md - Text and OCR → Semantic chunks: builds chunks
ReadMe.md - Semantic chunks → Chroma memory: persists chunks
ReadMe.md - Research notebook → PDF ingestion: ingests source
ReadMe.md - Research notebook → Evidence retrieval: runs retrieval
ReadMe.md - Research notebook → Grounded answer: generates answer
ReadMe.md - Research notebook → Visual panels: renders optionally
ReadMe.md - Research notebook → Brief export: exports optionally
ReadMe.md - Query rewriting → Evidence retrieval: rewrites queries
query_rewriting_pack/llama_index/packs/fusion_retriever/query_rewrite/base.py - Hybrid fusion → Evidence retrieval: fuses retrievers
query_rewriting_pack/llama_index/packs/fusion_retriever/hybrid_fusion/base.py - Chroma memory → Evidence retrieval: supplies memory
ReadMe.md - Evidence retrieval → Grounded answer: grounds response
ReadMe.md - Grounded answer → Model inference: requests inference
ReadMe.md