OpenAI, Qdrant, Cohere Reranker -- grounded RAG pipeline for messy legal documents — verbatim-quote-grounded first-pass memo + typed operator-edit learning loop. FastAPI + React + Docker.

This is the prompt i use day-to-day inside claude and ChatGPT web UI

Ehza's neovim preferences

A Codex CLI skill that pipes context to Claude Code for a second opinion.

Generic FAQ tag cleanup pipeline. Geometry-first (E5 + FAISS) with R1-R7 semantic-override rules and an advisory codex consultation step.

LLM-assisted FAQ corpus cleaner for tagged datasets — built for Bengali NID/voter (Bangladesh EC). E5+Gemma ranker, GPT-5.5 boundary policy + buffer audit, E5-only production filter.

Codex Chronicle — persistent, searchable engineering chronicles for OpenAI Codex sessions

Production-grade agentic chatbot for B2B customer service — domain-agnostic core with banking reference implementation

Persistent session knowledge tracker — captures planning decisions from AI coding sessions as durable markdown

Portfolio site for AI systems, computer vision, and agent tooling work

Claude Code Skill — get a read-only second opinion from OpenAI Codex

SwiftUI macOS media player for local video files

Download a Bengali Whisper Medium HF checkpoint (Kaggle or HF Hub) and convert it to CTranslate2 format for faster-whisper

Full-stack task management app built with Bun, React, SQLite, and JWT auth

SLM Reranker Benchmark: 7 models × 1,128 Bengali FAQ questions on T4 GPU