Everything here does the same basic thing: takes messy input and produces verified, automated output. Filter by what kind.
A RAG pipeline over mortgage loan documents. PaddleOCR extracts the text, FAISS handles retrieval, and answers come from a locally hosted Llama 3.1 with a swappable Gemini fallback — self-hostable with no external API key by default.
A 7-stage, zero-touch deployment pipeline. Jenkins triggers on GitHub Poll SCM, SonarQube enforces a custom quality gate, and Trivy scans for CVEs before anything ships.
A 4-stage B2B automation chaining four REST APIs. One domain goes in; verified decision-maker contacts and personalized outreach come out, with zero manual steps in between.
An XGBoost model trained on 15,000 e-commerce sessions (AUC 0.892), validated with an A/B test showing a 21.6% lift in conversion.
A compiler-design tool: a Flex-generated lexer and C logic identify the correct addressing mode across 8 types, wrapped in a small web frontend for interactive use.
93%+ accuracy telling real video frames from fake ones, using EfficientNetB0 fed by an automated OpenCV + MTCNN face-extraction pipeline.