Ingest
PDF / TXT / Markdown → extraction and chunks
Local RAG · Full-Stack AI Application / v1.0.0
A full-stack Retrieval-Augmented Generation application that turns PDF, TXT, and Markdown documents into a persistent, searchable knowledge base.
Separate TypeScript frontend and Python backend layers.
Documents are processed with page-aware extraction and overlapping chunks.
Retrieved chunks become context for llama3.2 with visible source metadata.
Index and metadata persist across backend restarts.
PDF / TXT / Markdown → extraction and chunks
Local embeddings → persistent FAISS
Question embedding → relevant context
llama3.2 → streamed answer + sources
This case study is based on the project’s own documentation.
Agentic AI · Production Engineering
A production-oriented agentic workflow platform with structured decisions, tool calling, human approval, asynchronous processing, and AWS Infrastructure as Code.
Local Speech AI · Full-Stack Application
A privacy-first meeting assistant that transcribes recordings locally, extracts structured notes, and answers grounded questions.
Evidence-Based AI · Full-Stack Application
A local-first CV-to-job matching assistant with transparent scoring, evidence states, and guarded AI writing.