Harsh Shaw
AI & Python Developer | LLM, RAG & FastAPI
New to Contra
Harsh is ready for their next project!
LLM Guardrails Gateway — Local LLM Security
Built a secure AI gateway using Python, FastAPI, Ollama, and Qwen2.5-Coder 7B. It protects LLM requests using PII detection, secret scanning, prompt injection and jailbreak detection, semantic safety checks, risk scoring, redaction, rate limiting, and audit logging. I also built a dashboard to show guardrail status, blocked threats, and request history.
0
28
A local RAG system that doesn't just retrieve and answer — it evaluates whether its own retrieval was good enough, and retries if not.
Upload a PDF, DOCX, or TXT file and ask questions against it. Instead of blindly trusting the first retrieval, the pipeline runs a critic/grounding stage that checks whether the retrieved context actually supports the answer — and if it doesn't, it rewrites the query and tries again (up to 3 times) before gracefully refusing.
LangGraph-orchestrated self-healing workflow: retrieve → generate → critique → rewrite → re-retrieve
Local embeddings (qwen3-embedding:0.6b) and generation (qwen2.5-coder:7b) via Ollama — fully offline inference
Persistent vector storage with ChromaDB
OCR fallback (Tesseract + pdf2image) for scanned PDFs
Transparent execution trace: similarity scores, retry count, grounding status, and rewritten queries all visible in the UI
Explicitly refuses to answer out-of-context questions instead of hallucinating
0
15
SpamShield is an ML-powered SMS and email spam detection application built with Python and Streamlit. I developed an NLP pipeline using NLTK for text preprocessing, TF-IDF for feature extraction, and Multinomial Naive Bayes for real-time spam classification. The application provides confidence scoring, message history, live session statistics, and an interactive dashboard for testing messages instantly.
0
21
Movie Recommendation System — Built a content-based movie recommender using Python, Pandas, Scikit-learn and CountVectorizer. It analyzes movie genres, cast, crew, keywords and descriptions to generate Top-5 similar movie recommendations, displayed through a Streamlit interface with TMDB posters.
0
34