Lia Martinez.
Back to selected work

Case 04 / Client work · Company anonymized

AI that prepares a firm’s required filings for submission.

In a forward-deployed engagement, I mapped a firm’s existing compliance process and designed AI to identify its filings, prioritize and assign them, prepare the forms from institutional memory, and automate the work up to human submission.

My role
Forward-deployed product owner and lead designer: process mapping, workflow automation, UX, and engineering specifications
Context
Multi-jurisdiction government-affairs compliance
Delivery
Designed; implemented by engineering

01 / The problem

Turn a firm’s existing process into an AI workflow

I worked from the firm’s actual compliance processes: what it had to file, how that work was organized, and who needed to be accountable. The goal was an operational tool that could carry those tasks forward, from identifying an obligation to preparing it for submission.

The system needed to determine the firm’s required filings, compile them in priority order, assign an accountable staff member, find the relevant forms, and prefill whatever the firm’s institutional memory could support. Automation would stop at submission, where a person had to take responsibility.

02 / My decisions

The choices behind the product

01

Start with the firm’s existing process

This was forward-deployed product work. I mapped how the firm handled compliance and used that process to define the AI workflow, rather than beginning with a generic filing checklist.

02

Connect obligation detection to priority and ownership

I designed the tool to identify the firm’s required filings, organize them in priority order, and attach an accountable staff member. Surfacing an obligation was the start of the workflow; the product also had to make clear who would move it forward.

03

Find the form and use institutional memory

The system finds the relevant filing forms and uses the firm’s accumulated information to prefill what it can. This connects identification to preparation and reduces the need to reconstruct information the firm already holds.

04

Keep unsupported information visible

Prefilling should rely on supportable information. Money fields remain empty when a value cannot be confirmed, and prepared values show whether they came from a record, a computation, or user input. The person can inspect the source before submission.

05

Automate up to the submission boundary

I designed automation to carry the filing through preparation and staging, with an explicit handoff for human certification and submission. The workflow supports the jurisdiction’s submission channel, while a person performs the final act.

06

Give the team several views of the same work

After five rounds of refinement, I rebuilt the experience around one obligation dataset shown as a status board, a deadline calendar, and a read-only legal ledger. These views support operational tracking without maintaining separate versions of the filing work.

03 / Delivery

What I delivered

After five rounds of refinement, I rebuilt the experience around one obligation dataset with a status board, deadline calendar, and read-only legal ledger. I delivered a functional prototype based on the firm’s mapped compliance process and researched against four jurisdictions. After CTO and founder review, engineering implemented it in production. I do not have measured filing-time or error-reduction results.

04 / The open question

What I’d test next

I would observe actual certifiers before treating the guardrails as effective. Source explanations only help if people use them. The unanswered question is whether reviewers verify the prepared values or simply click through.

About the evidence

Based on my anonymized AI portfolio and contribution record. This describes product design decisions, not legal advice or a claim of independently validated compliance.

Keep exploring

01 / Independent build · Work in progress

Teach your AI agents to work your way.

Read the case study
02 / Client work · Company anonymized

AI that uses a firm’s specialist knowledge to build strategy.

Read the case study
03 / Client work · Company anonymized

Turn specialists’ judgment into data an AI model can use.

Read the case study