Vela — AI Co-Pilot for Legal Research
How setren built an AI-powered legal research assistant that cuts case law analysis time by 70%. RAG pipeline, Claude integration, and a UI lawyers actually use.
The Challenge
Legal research is slow, expensive, and error-prone. Junior lawyers spend 60-80% of their time searching through case law databases, reading irrelevant results, and manually cross-referencing precedents. Vela's founding team — two former BigLaw associates — came to setren with a clear vision: build an AI co-pilot that understands legal reasoning, not just keyword matching. The tool needed to handle Indonesian and common law jurisdictions, cite sources accurately, and integrate into the workflow lawyers already use. Most critically, it had to be trustworthy — hallucinated case citations in legal work aren't just unhelpful, they're malpractice.
How setren Approached It
We started with a two-week discovery sprint, shadowing lawyers at three Jakarta firms to map their actual research workflow. This revealed that the bottleneck wasn't finding cases — it was understanding how cases relate to each other and to the client's specific situation. We designed a RAG (Retrieval-Augmented Generation) pipeline that indexes Indonesian court decisions, regulatory texts, and common law precedents. The retrieval layer uses hybrid search — semantic embeddings for conceptual similarity, plus BM25 for exact legal terminology. Every AI-generated answer includes pinned citations linking back to the source paragraph, so lawyers can verify in seconds. We chose Claude as the reasoning engine for its strong performance on multi-step legal analysis and its lower hallucination rate on factual claims. The frontend is a Next.js application with a split-pane interface: research query on the left, cited source documents on the right. We built it to feel like a conversation with a senior associate who has read every case.
The Result
Vela launched in Q1 2025 with three pilot law firms in Jakarta. Early results: lawyers report completing case law analysis 70% faster than manual research. The citation accuracy rate exceeds 94% — every generated reference maps to a real, retrievable document. The tool now handles over 200 research queries per day across the pilot firms. Vela's founders closed a seed round three months after launch, citing the product's traction and technical architecture as key factors in investor conversations.
“setren didn't just build what we asked for — they understood the legal domain deeply enough to push back on our assumptions and design something better.”
Tech Stack
Building an AI product?
We partner with founders to design, build, and ship AI-powered products. From RAG pipelines to production-grade LLM features — we've done it.
[email protected] →