FusionLogic: bringing operational intelligence to immigration law firms

Where it started: an Australian immigration law firm · Developed by: Vaxera AI · My role: Co-founder and Forward Deployed Engineer, Vaxera AI · Status: being deployed with three immigration law firms

Context

Immigration practices handle a high volume of detailed, deadline-driven work. Each matter involves applicants, often a sponsor, government lodgements, documents that expire, and a long trail of correspondence. A missed expiry or a document nobody followed up has real consequences for the client.

FusionLogic began as a forward deployed engineering engagement with one Australian immigration law firm. Instead of designing a product and then looking for a customer, the work started inside a practice, with its real data and its real staff.

The operational challenge

The firm has handled more than 5,000 matters. Like most practices, it kept its information in several places at once: a case-management system for the matter record, mailboxes for correspondence, a shared file store for documents, government portals for lodgement details, and accounting software for money. Each system did its job. None of them showed the whole picture.

So questions with simple answers took real effort. Which visas expire this month, across every client? What did we tell this sponsor in June? Which matters have gone quiet? Answering them meant opening several systems and assembling the answer by hand. And years of accumulated knowledge from past matters were almost impossible to search.

Approach

The work followed the same stages I use on any engagement.

• Understand and investigate. Time with the people running matters, and a close look at the firm's systems and data — including what the existing case-management system could and couldn't provide.
• Start with the knowledge. The first piece of work turned the firm's own files, correspondence and records into a searchable knowledge system, described in an earlier case study and three build notes.
• Shape it around real roles. Features were designed around how case managers, managers and the firm's accountant actually work, which is why each role now starts from its own home page.
• Privacy before access. Because the firm's data is confidential, a privacy gateway was built before staff were given an AI assistant that uses an external model.

What was built

Capabilities are grouped by their status today.

Built

• The matter record (Case Pilot) — The core of FusionLogic: matters, people, sponsors, notes, tasks and documents in one record, so FusionLogic can take over from a legacy case-management system rather than sit beside it.
• Matter at a glance — One page per matter that brings together the record, the people involved, checklists, documents and correspondence, in date order.
• Home pages by role — Case managers, managers and the accountant each start from what needs their attention.
• Deadlines and expiries — Visa and document expiries and follow-ups across every matter, sorted by urgency.
• Dashboards and business intelligence — The state of the practice: active matters, workloads and trends.
• Search and an AI assistant with sources — Keyword and meaning-based search across matters, correspondence and the firm's knowledge base, plus an assistant that answers in plain language and cites the documents it used.
• Case Story — An on-demand AI summary of a matter: its chronology, current position, outstanding items and dates, and suggested next steps (clearly labelled as suggestions), with references to the source documents. Each summary is saved and reused until the matter changes.
• Trust-balance visibility — Matters with trust balances and their movements, read from the practice's accounting data and shown beside the case record. This is a view of existing figures, not a trust-accounting system.
• Privacy gateway — Client identifiers are removed before any request reaches an external AI model, and staff can see on screen that this has happened.
• Works with existing systems — FusionLogic reads from a practice's case-management data, file store and accounting data rather than replacing them on day one.

In development

• Email capture — Correspondence filed to the right matter automatically, under rules the practice controls.
• Letters and forms from the record — Documents generated from a matter's details, with earlier versions kept.

Planned

• A client portal for progress updates, documents and questionnaires.
• Trust and office ledgers inside the application.
• An enquiries board, from first contact to an opened matter.
• Deeper integration with practice-management software, depending on what each vendor allows.

Technologies and architecture

• A web application with a Go back end and a Vue front end, Python services for practice data, document processing and the privacy gateway, and PostgreSQL.
• Search that combines keyword (BM25) and semantic retrieval with reranking, and local OCR so scanned documents are searchable too.
• An MCP server that gives the AI assistant typed tools over the firm's knowledge base, so answers come from the firm's own records and cite them.
• Everything runs on a server in the firm's own office, using local open-weight models where they fit and a frontier model through the privacy gateway where it adds value.

Current status

FusionLogic began inside one firm and is now being deployed with three immigration law firms. At its core is Case Pilot, the matter record built to take over from a practice's legacy case-management system, with the dashboards, assistant and privacy gateway working on top of it. It's still early, and the product is being refined with the people who use it.

What has been learned

Observations from the work so far, not measured results.

1. Real data finds what demo data never will. The firm's actual records held duplicates and inconsistencies that no test set contained. Finding them early, on real data, saved rework later.
2. Privacy has to be part of the architecture. The firm agreed to an AI assistant on confidential matters on one condition: client details are removed before anything leaves the firm, and that is shown on screen. Making privacy visible, not just promised, is what made the assistant acceptable.
3. Different roles need different starting points. A case manager, a principal and an accountant ask different first questions of the same data, so each now starts from a home page built for their role.
4. AI costs need limits from day one. Summaries are generated on demand, then saved and reused until a matter changes, rather than produced in bulk. That keeps costs predictable.
5. Data freshness depends on systems you don't control. How current the information is depends on how often data can be taken from existing software, so integration was a planning question from the start.
6. Sitting beside a legacy system has limits. Reading from the existing case-management system made the first version possible quickly. Making FusionLogic the place work actually happens needs a proper matter record — which is what Case Pilot is.

Next steps

• Complete the deployments under way with three immigration law firms.
• Keep refining FusionLogic with the people who use it.
• Bring the features in development — email capture, and letters and forms from the record — into client deployments.

---

Client details anonymised. FusionLogic is developed by Vaxera AI.