Student AI Digital Identity Core

Students accumulate valuable career milestones—resumes, certs, letters—across isolated files. Standard portfolios treat these as static attachments, forcing creators to manually type out descriptions, map skills, and maintain timelines. This creates massive friction, causing profiles to stay outdated. The gap is the lack of intelligent ingestion: a way to instantly convert raw academic text layers into structured, interconnected, and searchable career data. Process I engineered a decoupled full-stack architecture using a FastAPI backend and a Streamlit frontend UI. First, I set up a local PyPDF text layer extraction engine. I initially tried integrating external cloud AI services for parsing, but hit API quota limits. To build a robust, zero-cost solution, I shifted strategy and designed a custom, pure-Python local tokenizer and regex keyword anchor filter. This dynamically extracts skill tags, parses date markers, and categorizes documents into indexed SQLite tables. Finally, I connected the backend to a multi-tab Streamlit interface featuring visual metrics and metadata search engines. Results The platform completely eliminates manual profile assembly by enabling a 5-to-1 step workflow reduction—publishing portfolio items via a single drag-and-drop ingestion action. The zero-dependency local NLP layer achieves 100% classification automation without incurring cloud API costs. Next time, I would incorporate a direct visual graph visualization framework using a library like streamlit-agraph to replace the markdown text arrows. Reflection Instead of a text-based relationship mapper, I would integrate an interactive, visual graph rendering library like streamlit-agraph so users can physically drag and expand their career nodes. Additionally, I would implement a small, open-source local LLM (like Llama-3 via Ollama) to extract contextual summaries with deeper nuance than a pure-Python frequency algorithm can provide.