Inspiration
Many people experience a contract breach, financial loss or unfair treatment but do not know how to explain it to a lawyer, which facts matter or what evidence to preserve. The barrier is higher for people using a second language or living far from accessible legal services.
Lawyers meet the same problem from the other side. A first contact often contains memories, messages, contracts, receipts and emotional fragments rather than a coherent matter. The lawyer must identify the people, rebuild the chronology, separate assertions from evidence and locate missing material before substantive work can begin. During our tester recruitment, a lawyer described the cost of organising large collections of client material before the real legal work starts. That observation confirmed the need we are addressing.
AI Lawyer Opposition fills this missing preparation stage: it helps a person express the matter completely, organises the material, and lets professional time focus on verification, applicable law and strategy.
What it does
The main product is the NIDO StrikeOver dual-line lawyer workbench, not only a chat window. It organises parties, dates, events, objectives, both sides' positions, evidence and issues into one matter structure. It then expands review coverage across 18 factual, evidentiary, legal, procedural, causation, damages, burden-of-proof and strategic dimensions.
Two complementary review logics serve different stages. Open whole-matter discovery searches for contradictions, gaps and relationships that may not appear in the parties' existing positions. Contextual challenge places formed arguments back into the complete matter and tests their assumptions and evidence chains. Rapid review coordinates multiple angles into weakness cards. Advanced review lets each dimension reread the complete matter independently and preserves a full report, preventing one line of analysis from suppressing another.
Weakness cards retain their source dimension, model, evidence relationship and underlying report. They can be classified, moved into either side's argument area or sent into focused review. The workbench also supports two-round whole-case opposition, single-point attack and response, SWAP perspective reversal, and construction of likely attack methods from an opponent's objective.
The result is a lawyer-ready handoff package: intake record, background, parties, objectives, chronology, both positions, evidence index, claim-to-evidence links, contradictions, missing material, weaknesses, unresolved questions and lawyer review tasks. Five presentation modes cover a lawyer working paper, whole-case analysis, chronology and evidence index, weakness and evidence matrix, and client-readable review. Outputs include Markdown, PDF, DOCX and structured JSON.
Clients do not need to learn this complex workbench. A law firm can embed a simple reception window on its own website. Gemini asks focused questions about missing information, organises the response and preserves meaning across languages. The entrance also recognises document organisation, chronology, evidence indexing, PDF or Word formatting, OCR, bookmarks and pagination. With consent, the prepared request enters that firm's own workflow, allowing the firm to retain the client relationship while using the system for both intake and lead generation.
How we built it
We designed a staged professional workflow instead of one generic “analyse this case” prompt. The public layer is a simple web reception. Behind it is a Python-based lawyer workbench with structured matter objects, positive and negative side frames, evidence mapping, 18 review dimensions, weakness-card extraction, attack-and-response workflows and a shared professional report contract.
Gemini performs focused intake follow-up, matter organisation, position and evidence separation, report preparation and multi-angle weakness review. The deployed service runs on Google Cloud Run and uses Vertex AI. In our 5 August 2026 synthetic acceptance run, the desktop analysis used Gemini / gemini-2.5-flash. The online service extracted exact clauses from a synthetic contract PDF, organised both sides and completed a rapid 18-angle review. Cloud Run recorded a successful Vertex AI gemini-2.5-flash request.
Privacy is part of the architecture. Six operating routes cover online redacted or authorised original text, a local client with redacted or authorised AI assistance, fully local processing, and synthetic-case privacy mode. Sensitive material can remain local while fictional analogues are generated locally; only selected fictional material reaches an authorised model. The competition deployment keeps matter content in the active session rather than vendor-controlled long-term matter storage. Demonstrations use synthetic matters.
The report layer records report identity, stage, jurisdiction, provider, model, generation time, input scope, evidence, missing material and review tasks. Findings move from AI-generated and unverified, to lawyer review, to confirmed, modified or rejected. This turns model output into traceable professional working material.
Challenges we ran into
The first challenge was avoiding “just another legal chatbot.” Legal intake requires uncertainty, evidence relationships, missing-information tracking, opposing positions and a usable handoff—not merely a confident answer.
The second challenge was balancing depth with usability. Ordinary users need one simple entrance, while lawyers need independent multi-angle reports, adversarial review and professional exports. We solved this by separating the client-facing reception from the full dual-line workbench.
The third challenge was confidentiality. Persistent case storage would make demonstrations and analytics easier, but legal trust requires restraint. We chose privacy-safe sessions, synthetic demonstrations and local processing routes.
Accomplishments that we're proud of
We built a working public intake flow connected to a much deeper lawyer-side system. The five offline professional report modes and Markdown, PDF, DOCX and JSON generation passed synthetic-data tests. The online flow successfully extracted contract terms from a PDF, built structured party materials and completed an 18-angle review through Gemini on Vertex AI.
We also created a reproducible market-demand snapshot. As of 1 August 2026, it documented 15 public projects requesting legal intake, document processing, OCR, evidence organisation, chronology or AI litigation workflows. Disclosed fixed budgets totalled approximately A$9,065–A$10,026. Australia has 16,793 private law practices, 78% of them sole practices, providing a clear first market for subscriptions, deployment and private-environment support.
What we learned
The largest opportunity is not simply that AI can answer legal questions. It is workflow transformation. Clients need a safe way to explain a matter; lawyers need a prepared record that helps them understand it quickly. Gemini is most valuable between those two needs: asking better questions, organising material, mapping evidence and widening the set of issues presented for professional review.
What's next for AI Lawyer Opposition
Gemini was also used to assist with the preparation and adversarial review of the founder's patent application. The application concerns a multi-task AI processing method designed to support large-scale multi-round reasoning, substantially reduce token consumption and mitigate hallucination risk through independent verification. During preparation, Gemini helped identify possible omissions, ambiguous claim boundaries and potential design-around paths through multiple rounds of adversarial testing.
Our next product steps are stronger firm-controlled accounts and retention policies, clearer deployment packages for local, private-cloud and Google Cloud environments, and jurisdiction-specific professional workflows. Our goal is to turn scattered information into checkable structure and reserve scarce legal expertise for the work that most needs human experience and judgment.
Copyright © 2026. All rights reserved. Competition review and authorised testing do not grant permission to copy, redistribute, reverse engineer or commercially use the software, workflow, documentation or demonstration materials. Patent applications have been filed for the underlying multi-task processing architecture; public materials describe product outcomes without disclosing unpublished implementation details.
Log in or sign up for Devpost to join the conversation.