oluwafemidiakhoa/gaialab-naija-assistant

# GaiaLab Naija Assistant GaiaLab Naija Assistant is an open-source AI research project focused on practical language models for Nigerian communication, small-business support, customer service, professional writing, and locally relevant assistance. ## Current Release **GaiaLab Naija Assistant v0.3** - Base model: `Qwen/Qwen2.5-0.5B-Instruct` - Training method: LoRA - Total reviewed records: 212 - Training records: 170 - Validation records: 42 - Epochs: 3 - LoRA rank: 16 - LoRA alpha: 32 - Final validation loss: 1.305 - Status: Experimental research release ## Project Goals - Nigerian small-business communication - Customer-service responses - Professional emails and WhatsApp business messages - Nigerian English and Nigerian Pidgin - Delivery and payment updates - Agriculture-related assistance - AI literacy and scam-awareness guidance ## Repository Structure ```text app/ Application files config/ General project configuration data/ Training and evaluation datasets evaluation/ Human review and benchmark tools notebooks/ Research and experimentation notebooks outputs/ Training and evaluation outputs scripts/ Data preparation and utility scripts src/ Core source code tests/ Automated tests training/ Training configurations ``` ## Dataset Management The local dataset platform provides immutable version snapshots, append-only human review, per-example SHA-256 hashes, semantic-duplicate reports, automatic statistics, CSV/JSONL releases, and cross-model benchmark summaries. See docs/dataset_management_platform.md for the reproducible workflow. Launch the reviewer with: ```bash streamlit run app/dataset_review.py ``` Public, privacy-preserving release verification is documented in docs/public_release_verification.md. ## Governance platform The production governance layer adds deterministic advisory quality assessment, role-aware append-only human review, training eligibility, release scorecards, write …