This repo tackles one of the messiest problems in applied AI: turning a pile of regulatory PDFs into something a language model can actually reason over. It's a legal document ingestion pipeline targeted at Indian energy regulators — CERC (Central Electricity Regulatory Commission), CEA (Central Electricity Authority), and MNRE (Ministry of New and Renewable Energy) — but the architecture is general enough to be interesting well beyond that niche.

The pipeline does four things end-to-end:

• Automated collection — scrapes and pulls regulatory PDFs directly from agency sources, so the corpus stays fresh without manual downloading.
• Layout-aware parsing — regulatory documents are notoriously ugly (multi-column layouts, embedded tables, scanned pages, footnotes that matter legally). Getting this right is 80% of the battle.
• Dynamic categorization — organizes documents by legal category rather than dumping everything into one flat index, which is critical for retrieval precision when a single query might otherwise pull tariff orders alongside safety regulations.
• Vector database construction — builds semantic embeddings tuned for legal research queries, forming the retrieval layer for a downstream RAG system.

What makes this interesting is the production-grade framing. Most RAG tutorials stop at "chunk your PDFs and embed them." Real legal work needs jurisdictional awareness, category-aware retrieval, and parsers that don't silently mangle a table of tariff rates. If the implementation delivers on the description, it's a useful reference for anyone building domain-specific RAG in regulated industries — healthcare, finance, compliance.

Who benefits: Legal-tech developers, energy sector analysts working with Indian regulatory bodies, and RAG builders looking for a concrete example of vertical-domain retrieval that goes beyond the toy examples. Even if you're not in energy law, the layout-parsing and dynamic-categorization patterns are worth studying.