Portfolio
ShayarAI: Transformer‑Based Urdu Shayari Generator (Style, Rhyme, and Meter‑Aware)
An end-to-end Urdu poetry generation system: ingestion/cleaning, Urdu tokenization, GPT-style Transformer fine-tuning, style/topic conditioning, and meter/rhyme evaluation hooks for responsible creative AI.
MuFaw AI Research LabNLPTransformersUrduCreative AITokenization
Project details
What we delivered
Overview
Generates original Urdu couplets using a GPT-2-like Transformer trained on structured poet/verse pairs, with explicit AI labeling and guardrails against misattribution.
Architecture
Ingest public poetry → Urdu-specific cleaning → dataset [Poet, Couplets] → Urdu BPE tokenization → fine-tune Transformer → generate with style/topic prompts → evaluate with meter/rhyme heuristics.
Responsible Deployment Notes
Respect source licensing/policies.
Clearly label outputs as AI-generated.
No claims that generated text is by real poets; only style-conditioned inspiration.