Pengacara-ku — legal research engine for the Indonesian corpus.
Hybrid retrieval across 18,000+ Indonesian court rulings with explicit citation to ruling numbers — built for lawyers, paralegals, and legal academics who need research in seconds without hallucination risk.
Tens of thousands of Indonesian court rulings are scattered across lengthy PDF documents. Traditional keyword search fails to accommodate legal synonyms, spelling variations, and OCR noise. Lawyers spend hours — even days — locating relevant reference rulings. Pengacara-ku compresses the research cycle to seconds, with every answer designed to cite the original, verifiable ruling number.
Knowledge-intensive domains such as Indonesian law require an AI that is willing to say "not found" rather than fabricating an answer — because hallucination in a legal advice context can have fatal consequences for a client.
Five-stage pipeline.
Each stage is designed to minimise hallucination and guarantee traceability to the original ruling.
- 01
Ingestion
Court rulings in PDF format are ingested into object storage; metadata is extracted to a relational database indexed by ruling number, year, and case type.
- 02
Chunking
Long documents are broken into 262,000+ semantic chunks with maintained overlap — each chunk carries its source ruling metadata, guaranteeing traceability down to paragraph level.
- 03
Dual embedding
Each chunk is indexed on two representations: a keyword index (FTS5) for exact search and a vector embedding for semantic search. The combination mitigates the weaknesses of each approach individually.
- 04
Hybrid retrieval
User queries are run in parallel across FTS5 and vector search; results are merged with confidence-score re-ranking. The threshold is set explicitly: below it, the system returns "not found".
- 05
RAG with citation
The LLM composes answers from retrieved chunks, with system instructions to always cite the ruling number. Every claim is verifiable by the user against the original source in one click.
Scale and traceability.
Pengacara-ku has been running in production since early 2025. The corpus grows continuously as the Supreme Court publishes new rulings. Every query produces an answer with full citation to the ruling number — a feature required for an AI system to be accountable in a legal practice context.
p95 latency is under three seconds for retrieval search, and under eight seconds for deep-analysis chat.
Choose the right plan.
Subscribe from day one — from Rp 39,000/month for law students to the Firma plan for office teams.
Individual
For law students, lawyers, paralegals, and in-house legal teams.
- Kampus Rp 39,000/month — law students, read access
- Fokus 1 bidang Rp 149,000/month
- Fokus up to 3 bidang Rp 199,000/month
- Praktik Lengkap Rp 249,000/month — full access
Firma & Enterprise
For law firms, in-house legal teams, and large institutions.
- Firma Rp 249,000/month — 2 to 5 members
- Enterprise Rp 999,000/month — coming soon
- Multi-user access
- Dedicated support & priority uptime SLA
Questions frequently asked.
Clarification on questions commonly raised by Pengacara-ku users.
What is Pengacara-ku? [ + ]
Pengacara-ku is an AI-powered legal research platform that makes it easy to search and analyze Indonesian court rulings using natural language.
Where does the ruling data come from? [ + ]
Ruling data is sourced from the Supreme Court of the Republic of Indonesia’s official Decision Directory — which covers decisions from religious, district, high, administrative, military, and Supreme Court levels — and processed for more efficient retrieval.
Can it be used for academic research? [ + ]
Yes, Pengacara-ku is well-suited for academic research, theses, dissertations, and day-to-day legal practice. Each ruling includes complete metadata for citation purposes.
Pengacara-ku is an implementation of the hybrid RAG pattern for the Indonesian legal knowledge domain. A similar pattern can be built for your organization's specific domain — corporate contracts, internal regulations, claim documents, operational knowledge bases, or team workflows that consume significant time.
The engineering implementation is structured through a Pilot Build — a contract engagement with measurable deliverables, observability, audit trail, and fail-safes to production-grade standards.
Learn about Pilot Build engagementTry Pengacara-ku now or discuss a Purpose-Built AI Agent for your organization.
From Rp 249,000/month. For engagement patterns, the WTB engineering team opens thirty-minute discussion sessions.