AI Systems Engineer | Autonomous Agents · RAG · Retrieval Infrastructure

devto

github

Qasim Sethar  • Peerlist

Central hub for production-style AI projects including semantic search, document ingestion pipelines, and LLM retrieval architectures. Live: https://qasimio.github.io/

Applied AI & Retrieval Systems Portfolio

FOLDR is a lightweight CLI tool that instantly organizes messy directories by categorizing files into clear folders based on extension. Built with safety first, it never modifies existing directory structures and includes a --dry-run feature to preview all changes before any file is moved. Perfect for predictably cleaning up Downloads folders.

FOLDR

MQNotebook — Enterprise RAG System Python, LlamaIndex, OCR, Hybrid Search Built a local RAG engine ingesting scanned PDFs, PPTX, XLSX with custom OCR pipeline. Implemented hybrid sparse/dense retrieval with cross-encoder reranking, improving context precision by 40%. Reduced token usage by 60% via optimized context injection and chunking strategy. Deployed to Streamlit Cloud with BYOK security.

MQNotebook

DevShelf is a high-performance distributed search engine built specifically for Computer Science textbooks and academic material. Written entirely in Java, it delivers fast, precise retrieval using a custom inverted indexing system and vector-space ranking designed for near-constant-time lookup. Unlike traditional search stacks, DevShelf is built from the ground up without Lucene or Elasticsearch. This keeps the engine lightweight, transparent, and fully controllable while still supporting advanced retrieval techniques such as TF-IDF scoring, semantic vector ranking, and behavioral relevance signals. The system follows an offline-first architecture, allowing local usage with zero cloud dependency while still supporting optional cloud-synchronized index updates for continuously refreshed data. Its distributed design enables metadata and index streaming from serverless sources, ensuring users always access the latest searchable content. DevShelf is fully open source and structured using Domain-Driven Design principles. The architecture emphasizes performance, modularity, and extensibility, supporting features such as recommendation graphs, dynamic filtering, and both CLI and JavaFX graphical interfaces. Designed for students, researchers, and developers working with large academic corpora, DevShelf provides a fast, transparent, and developer-friendly alternative to heavyweight search frameworks. Open to freelance search, retrieval or backend engineering collaboration.