AI-powered annual report generator for Singapore SMEs

The Challenge

Annual report preparation is one of the most time-expensive workflows in Singapore accounting. Corporate Service Providers spend days per engagement mapping accounts to ACRA taxonomy, drafting SFRS-compliant disclosures, and assembling notes by hand. Existing tools either solve a sliver of the work or carry a Big Four price tag.

Our Approach

Account mapping is the bottleneck of every annual report engagement, so we put AI at the core of it. The mapping engine matches each general ledger account to its ACRA taxonomy tag using vector embeddings against a curated taxonomy index, with confidence-gated LLM disambiguation only when the embedding signal isn't decisive. A 187-account ledger maps in under 5 minutes at 95%+ confidence, and every confirmed mapping makes the next engagement faster. Notes and accounting policies assemble from a versioned template library so the same input always produces the same output. The end product is a single board-ready Word document, ready for AGM signing and ACRA submission.

Where AI does the work

• Embedding-based ACRA taxonomy mapping with a pgvector-indexed taxonomy and Voyage embeddings for semantic account-to-tag retrieval.
• Confidence-gated Claude disambiguation that fires only when retrieval signal isn't decisive, keeping LLM cost and latency low.
• Cross-engagement learning that caches confirmed mappings across the firm so repeat patterns map instantly.
• Provenance tracking on every mapping (prior-year, embedding, AI-suggest, user-override) so the audit trail is replayable.