newcase.ai

**We built it in 10 days. Put it in front of four litigation attorneys, including two managing partners with 100 years of combined experience. By the end, they were asking to use it on real cases.** In multimillion-dollar litigation, an expert witness can make or break a case. To challenge one, attorneys hunt for contradictions across years of testimony. The litigators showed us what that entails: weeks navigating outdated court databases, purchasing filings behind paywalls, and reading hundreds of pages. **The cost? over 20 hours of manual work, costing over $5,000 per reserach.** Worse, the most valuable evidence may be in a case they don't know exists. One filing reveals a new lead, another court, another paywall. Research becomes a branching investigation that traditional search cannot complete. **During this hackathon, we built an autonomous expert-witness investigator from scratch.** The hard part wasn't prompting an LLM. It was enabling AI to do the work. Court records span 50+ jurisdictions, with fragmented indexing, separate credentials, legacy interfaces, and paid access. We engineered a **purpose-built agent harness** with tools to search legal sources, navigate court records, retrieve and parse filings, and request paywalled documents. Our payment layer enforces spending limits and purchase rules, with support for single-use virtual cards, so AI can pursue evidence with financial safeguards. **Meta's Muse Spark 1.3 is the decision-making engine.** Through its OpenAI-compatible API, we integrated Muse into **Google ADK via LiteLLM**, orchestrating a lead investigator and specialized research sub-agents. Muse chooses tools, evaluates results, delegates court searches, and recursively follows promising leads. We integrated **Meta's native web search through the Responses API** to uncover new cases with source citations. Its speed makes repeated reasoning and tool calls practical; structured outputs and evidence-validation checks help keep findings grounded in source documents. **We built with attorneys, not just for them.** We shipped the agent to four practicing litigators during the hackathon. Two were managing partners. They challenged our findings, revealed overlooked data sources, walked us through their research workflows, and showed us what evidence attorneys actually need. We incorporated their insights, improved the investigation workflow, and put the results back in front of them. **The result:** attorneys who ordinarily spend thousands on this research asked to use our agent on their own cases. In a litigation system with trillions of dollars at stake, uncovering buried evidence can change case outcomes. **In 10 days, we built, shipped, learned, iterated, and validated demand for an autonomous investigator attorneys want to use now.**