Case study · Legal
"Practice Made Perfect": a full agent platform a partnership could see.
Law firms don't buy futures. They buy tangible things partners can use and clients can trust. So before scoping months of engineering, we built a branded, working reference platform with seven integrated modules, Assistant, Vault, Knowledge, Workflow Agents, ROI Analytics, Security & Governance, and a Quantilus Academy, and walked the partnership through it.
integrated platform modules
privileged-data deployment model
built into the dashboard, not a slide
partnership sign-off to proceed
The problem
A partnership cautious about AI, surrounded by vendors promising magic.
A mid-market law firm had heard every AI pitch in the industry, from feature-bolt-ons in their existing practice management system to standalone "legal AI" point products. The partnership wanted to move, but every concrete commitment had hit the same wall: where does our privileged data go, who controls it, and what do we actually see when we open the application?
The marketing director and managing partner brought us a hard but reasonable ask: before approving real budget for a custom build, they wanted a tangible reference that went beyond slides and Figma prototypes. Something that looked like the product they'd run, addressed the privileged-data concern up front, and made the ROI story explicit instead of implied.
The approach
Build the reference platform first. Scope the engineering second.
We invert the usual sequence. Most consultancies start with a discovery deck and end with a SOW. We started with a deployable, branded, working reference platform, "Practice Made Perfect", that the partnership could click through, demo internally, and react to before we wrote a line of production code. The seven modules of the platform map to the seven workflows the firm cared about most, and the platform itself became the spec for the build.
Module 01 · Assistant
A partner's daily co-pilot, scoped to their practice.
Surfaces relevant matters, deadlines, billing reminders, and inbox triage in the partner's voice. Answers internal Q&A from the firm's own knowledge base. Knows enough about the matter at hand to draft initial responses without being asked twice.
Module 02 · Vault
Document repository with semantic search and lineage.
Every document, motion, contract, brief, memo, with full-text + semantic retrieval. Privileged-marker enforcement at the row level. Lineage tracking so a partner can see how a clause moved through a deal, every diff, every signer, every comment.
Module 03 · Knowledge
The firm's accumulated expertise, queryable.
Internal precedent, prior matter outcomes, expert opinions, redacted briefs from analogous matters. The agent finds the right people and the right prior work in the firm before reaching for external sources.
Module 04 · Workflow Agents
Specialized agents for matter intake, billing, due diligence.
Discrete agents handling the long tail: new-matter intake triage, conflict-check support, redline-summarization, billing-narrative drafting, deposition-prep digesting, e-discovery sorting. Each agent has a defined scope, a defined approval gate, and shows up in the same UI.
Module 05 · ROI Analytics
"What did the platform earn this month?"
Hours saved per matter, draft acceleration, billable-time recovered, intake-triage backlog reduced, measured against a baseline the firm sets at install. The number lives in the platform; partners don't have to ask their finance team to prove the platform is working.
Module 06 · Security & Governance
"Where exactly does the data go?", answered visually.
A live view of which models the firm is using (open-weight in-VPC, frontier-via-cloud-account, or air-gapped), which matters use which model, who approved each policy change, and who accessed which privileged document. Auditors love it. Partners trust it.
Module 07 · Quantilus Academy
Training the partnership instead of replacing it.
Short, role-specific learning paths inside the platform, "how to use Vault as a litigator," "how to read the ROI dashboard as a managing partner," "how to set policy boundaries as the IT director." The platform comes with the change-management built in, instead of as a separate engagement.
Architecture
Reference build first, production build informed by it.
- Reference platform. Static HTML + vanilla JS + custom CSS, branded as "Practice Made Perfect." Designed to look like the real product, click like the real product, and answer the partnership's real questions about flow and posture, without taking on the cost of a production build before approval.
- Privileged-data posture. Open-weight LLMs in the firm's AWS VPC for matter-level work, frontier models via Bedrock for non-privileged research, fully air-gapped option available for the partnership's most sensitive matters. Picked early and surfaced in the Security & Governance module so reviewers can see the model gateway clearly.
- Integration surface. Designed for connectors to iManage / NetDocuments (DMS), Elite 3E / Aderant (billing), Intapp Open / Litera Transact (intake / workflow), and Outlook / Gmail for inbox triage.
- Deployment plan. Reference platform on a static hosting environment for partner walkthrough; production build runs in the firm's own AWS account with SOC 2 controls, KMS-managed keys, and per-matter audit trails.
The result
Stakeholder buy-in before we wrote production code.
The reference platform did what it was built to do: a partnership that couldn't agree on whether they wanted AI walked out of the demo agreeing on what they wanted AI to look like inside the firm. The seven modules and the Security & Governance view in particular collapsed weeks of "how does this fit our compliance posture?" conversation into a single live demo.
The downstream effect was material: the production engagement was scoped against the reference platform's specific UI flows, which meant the SOW landed faster, the procurement review went smoother, and the production build started with a partnership already aligned on what success looked like.
Lesson worth taking: for risk-averse buyers, law firms, hospitals, banks, the cheapest piece of work you can do is build the thing that lets them react to a real artifact. Slides and demo videos lose to a clickable platform that addresses their real concerns up front.
Tech stack
Reference build
Static HTML / CSS / vanilla JS, custom design system, deployed to S3 + CloudFront for partner walkthrough
Planned production
Open-weight LLM in firm AWS VPC, AWS Bedrock for non-privileged work, Postgres + pgvector, audit-log-first architecture
Integration targets
iManage / NetDocuments DMS, Elite 3E / Aderant billing, Intapp / Litera workflow, Outlook / Gmail, custom internal taxonomy APIs
Patterns worth stealing
Three lessons from this engagement.
- For risk-averse buyers, build a reference before you scope a build. A clickable reference platform is the cheapest piece of code you can write to collapse months of "what would this even look like?" debate.
- Make the data-residency story visible. The Security & Governance module didn't just exist, it lived on the home screen. For privileged-data buyers, that visibility is the deciding factor.
- Bake ROI into the product, not the sales deck. When the dashboard shows what the platform earned this month, partners stop asking. The number defends itself.