This bounty is for the iOS version of Kenny’s Best Offline Ai Research app. It loosely follows the same requirements and structure but is specifically for the iOS platform. (Same requirements in a iOS compatible way if funded by others.)
Referenced bounty below
Offline Research iOS — offline FTS5 retrieval + on-device GGUF, cited answers
Repo: https://github.com/ciperx/offline-research-ios — SwiftUI iOS 16 app: bundled SQLite FTS5 corpus (115 articles / 1031 chunks / 4.3 MB) + hybrid BM25/lexical retrieval + extractive answers with inline citations + optional llama.cpp GGUF engine behind an LLMEngine protocol. Zero network APIs in the app (CI deny-list + URLProtocol block + audit counter). Verified on Linux with Swift 6.0.3: swiftc -parse 33 files clean, -typecheck 13-file core, retriever smoke test 6/6 on the real index, retrieval eval 53 questions hit@3 0.925 / MRR 0.881. NOT verified: Xcode build, xcodegen generate, llama.cpp compile, device performance — docs/BENCHMARKS.md has the protocol. MIT code, CC BY-SA 4.0 corpus with per-article attribution.
Offline Research iOS — offline FTS5 retrieval + on-device GGUF, cited answers
Offline AI Research for iOS - on-device RAG with citations
Swift/SwiftUI app + OfflineResearchCore package. Local SQLite FTS5 corpus 369 docs / 17,846 chunks; hybrid BM25+lexical ranker with RRF; extractive answers with inline citations; optional llama.cpp GGUF engine; zero network in research path. Eval hit@5 0.588 MRR 0.495.
OfflineResearch iOS - offline AI research app
Offline iOS research app for poidh #1405: on-device SQLite FTS5 knowledge pack (116 articles), cited extractive QA (explanation/comparison/synthesis/factual), 0 network bytes in the research path (proven by sockets-blocked unit test, 17/17 pass).
September 19, 2026
POIDH (Pics Or It Didn’t Happen) is a decentralized bounty platform where anyone can create challenges and reward proof with ETH.