Offline NLP Study Assistant
Budget: $250 – $750 USD
I have roughly 1,600 pages of course notes sitting in a single Google Doc and I want them to become an always-available study companion. The job is to turn that material into an offline, text-question / text-answer assistant that can pull the right information, cite where it found it, and run entirely from my laptop without needing an internet connection.
What matters to me
• Natural Language Processing is the core: clean question understanding, precise retrieval, and clear, human-readable answers.
• All answers must be generated from my notes only—no hallucinations or web look-ups.
• The full note set will be stored locally; assume zero connectivity once the system is installed.
• Responses are strictly text; no audio or diagrams are required right now.
Key deliverables
– A preprocessing script that exports the Google Doc, segments the content sensibly (chapters, headings, paragraphs), and builds a local, searchable representation (embeddings, TF-IDF, or comparable NLP indexing of your choice).
– A lightweight inference engine—Python preferred—that accepts a user question, runs it through the local index, and returns a concise, referenced answer (e.g., “According to Week 3 ‑ Linear Algebra, p. 27…”).
– A minimal interface (CLI or simple desktop/web GUI) so I can start the assistant, type a question, and read the reply.
– Installation instructions plus a short README so I can regenerate or update the index when I add new notes.
Acceptance criteria
1. Works with Wi-Fi disabled.
2. Answers are drawn from the notes with at least one inline citation.
3. Average response time under five seconds on a mid-range laptop.
4. Setup can be reproduced from scratch using only the provided README.
If you have experience with local LLMs, embeddings, or libraries like spaCy, sentence-transformers, LangChain, or Chroma, please mention it in your bid along with an outline of how you would approach the indexing and retrieval steps.
What matters to me
• Natural Language Processing is the core: clean question understanding, precise retrieval, and clear, human-readable answers.
• All answers must be generated from my notes only—no hallucinations or web look-ups.
• The full note set will be stored locally; assume zero connectivity once the system is installed.
• Responses are strictly text; no audio or diagrams are required right now.
Key deliverables
– A preprocessing script that exports the Google Doc, segments the content sensibly (chapters, headings, paragraphs), and builds a local, searchable representation (embeddings, TF-IDF, or comparable NLP indexing of your choice).
– A lightweight inference engine—Python preferred—that accepts a user question, runs it through the local index, and returns a concise, referenced answer (e.g., “According to Week 3 ‑ Linear Algebra, p. 27…”).
– A minimal interface (CLI or simple desktop/web GUI) so I can start the assistant, type a question, and read the reply.
– Installation instructions plus a short README so I can regenerate or update the index when I add new notes.
Acceptance criteria
1. Works with Wi-Fi disabled.
2. Answers are drawn from the notes with at least one inline citation.
3. Average response time under five seconds on a mid-range laptop.
4. Setup can be reproduced from scratch using only the provided README.
If you have experience with local LLMs, embeddings, or libraries like spaCy, sentence-transformers, LangChain, or Chroma, please mention it in your bid along with an outline of how you would approach the indexing and retrieval steps.