PROJECT

Email ingestion: turning the inbox into a system of record

How I designed email ingestion for an AI-native CRE lending platform, from PRD to production in nine weeks.

Fintech

AI Product

B2B SaaS

Platform Primitive

cliff notes

Overview: At Hypha, I designed email ingestion, the system that lets commercial real estate lenders forward or CC any deal email and have its body and attachments captured, routed to the right entity, and preserved as citable provenance. No download/upload step, no leaving the inbox behavior they already have.


Challenge: Deal documents are born in email, offering memos from brokers, covenant packages from borrowers, executed agreements from counsel, but reaching Hypha required someone to notice, download, and re-upload them by hand. Every ingested email then demanded three routing decisions (organization, entity, document handling), and the design had to work as the fallback for an agent expected to automate routing 90% of the time.


Solution: A selection-first routing model where the whole email package files in one click, a suggestion state built to hand off to the agent, an ephemeral action-oriented surface instead of a second inbox, and a provenance chain that lets any extracted value cite the exact email it came from.


Impact: Shipped September 25, 2026: live with three customers on day one, built as a flexible primitive that other teams now layer deal creation and attribute versioning on top of. A filing workflow of notice, download, re-upload, and categorize per document became one click per email.

DATE

July-September 2026

MY ROLE

Product Designer

TOOLS AND TECH

  • Figma

  • FigJam

  • Claude AI

  • Linear

CORE TEAM

1 Product Designer

1 Product Manager

1 Front End Engineer
1 Full Stack Engineer

A peek into the impact of this project:

From PRD to production

0

weeks

Customers

0

on day one

Routing

0

%

automated

Business objectives

Connecting design to the business strategy

Hypha’s core bet is being the system of record for CRE loan data: every time a user leaves Hypha to move a document, the value proposition weakens. And in this industry, deals are born in email. The immediate driver was our deal analyzer work, where an originator told us directly: if he could CC Hypha on broker threads and have correspondence stay with the deal, that would be the glue keeping Hypha his source of truth.

Capture without behavior change:
Users forward or CC an address they already know how to use. The market pattern existed (Ramp proved forward-to-ingest for expense management), but nobody had built it for CRE lending.

Capture without behavior change:
Users forward or CC an address they already know how to use. The market pattern existed (Ramp proved forward-to-ingest for expense management), but nobody had built it for CRE lending.

Make three decisions feel like one:

Every email needs an organization, an entity, and document handling resolved. The common case had to be a single click.

Make three decisions feel like one:

Every email needs an organization, an entity, and document handling resolved. The common case had to be a single click.

Preserve provenance:
Knowing "what the seller actually sent, and when" was a named customer pain. Captured email had to stay citable, not dissolve into anonymous files.

Preserve provenance:
Knowing "what the seller actually sent, and when" was a named customer pain. Captured email had to stay citable, not dissolve into anonymous files.

Design the fallback that teaches the agent:
Routing was expected to automate "90% of the time." The manual UI had to be the training interface for that future, not something the agent would replace.

Design the fallback that teaches the agent:
Routing was expected to automate "90% of the time." The manual UI had to be the training interface for that future, not something the agent would replace.

Step 1: define the problem

Understanding the core pain-points

The current experience created an additional step for users

Documents arrived scattered across inboxes with no capture step. Someone had to notice an email, download its attachments, and re-upload them by hand before any of Hypha’s document intelligence could touch them.

User Problem

CRE lenders and originators receive deal documents by email (from brokers, borrowers, sellers, and counsel), but have no reliable way to capture them, route them to the right deal, or later prove what was actually sent and when. The result is manual re-entry, incomplete file sets, and diligence that can’t tell what came in versus what didn’t.

Step 2: research and discovery

Starting n a working build, not a blank canvas

Starting n a working build, not a blank canvas

Starting from a working build, not a blank canvas

Consumer

Uncertain, non-expert, time-poor

Needs clarity on their situation and a clear next step. AI should reduce any confusion.

  • Understand tax picture instantly

  • Directed, prioritized actions

  • Trust in their advisor

Tax Professional

Skeptical, expert, deadline-driven

Needs portfolio visibility, fast docs, and proactive signals. AI should augment judgment.

  • Season-level portfolio view

  • Automated document intake

  • Proactive client flags

Step 3: design strategy

Rethinking the information architecture

Rethinking the information architecture

The MVP had no navigation logic. I restructured both platforms from scratch around user intent.

The consumer nav was reorganized into three distinct groups based on what users are trying to do: understand their situation, take action, and manage their tax picture. The tax pro nav was organized around the advisor workflow lifecycle from prospecting to filing. Both platforms share one AI layer and one component library underneath.

Consumer IA decisions

Grouped by user intent: understand your situation, take action, manage your tax picture. The AI bar persists across all sections because it is always relevant.

Tax Pro IA decisions

Organized around the advisor workflow lifecycle. Prepare is the most complex section and gets its own sub-workflow: Gather, Review, Approve, File.

Step 4: ai ASSISTED wireframing

Using AI in the Design Process

Technical requirement documents are written for engineers. I use Claude to translate them into design decisions.

I use Claude to parse dense technical specs: extracting data inputs and outputs, identifying what the frontend needs to collect and display, and flagging UX implications buried in backend logic. That parsed understanding feeds directly into a rough wireframe, giving me something to pressure-test before investing in high-fidelity work. The result is a tighter loop between technical constraints and design decisions, and less back-and-forth with engineering on fundamentals.

Step 5: collaboration

Building in Partnership With Engineering

Designing for a tax platform means the front end and back end are deeply connected. Before and throughout the design process, I worked closely with engineering to make sure nothing fell through the cracks.

Technical Requirements Review: I read through engineering's technical requirement documents to understand what data the backend needed to collect and how it had to be structured. This ensured the frontend was capturing the right information at the right time

Translating Backend Logic to the Front End: Tax data is complex under the hood. When information from the backend needed to be surfaced to consumers, I made sure it was displayed in a way that was clear and actionable, not just technically present

Pressure-Testing Ideas Early: Before moving into full designs, I ran concepts by engineering to identify feasibility issues early. This kept the design process moving without pivots later

step 6: designing the consumer experience

step 5: the consumer experience

Designed for clarity amidst anxiety: the consumer dashboard

Designed for clarity amidst anxiety: the consumer dashboard

The consumer dashboard

The Problem:

No financial visibility: High-net-worth consumers filing complex returns had no way to understand their current tax situation, outstanding obligations, or expected outcomes

Minimal MVP foundation: The existing experience offered only a welcome banner and a document list, with no status indicators, financial context, or sense of progress

No AI value surface: There was no mechanism to surface AI-generated insights, leaving the product’s core differentiator completely invisible to the end user

The Solution:

Client-centered dashboard architecture: Organized the dashboard around the four questions every client asks at the start of tax season, mapping each directly to a KPI card with year-over-year trend context

Prioritized action feed: Surfaced exactly what each client needs to act on next, with time estimates per task to reduce friction and set clear expectations

AI insights panel: Designed a dedicated panel that surfaces savings opportunities, risk flags, and wealth moves with estimated dollar impact per recommendation

Advisor conversion mechanism: Positioned the insights panel as the core business driver, translating client engagement into scheduled advisor actions

The structure view

The Problem:


Consumers managing multiple LLCs, partnerships, and trusts had no way to understand how their entity structure fit together or how each entity affected their personal return.

The Solution:

Entity hierarchy visualization: Displays the full ownership structure, filing type, and status for each entity in a scannable org view and list view toggle

Immediate sophistication signal: Communicates OLarry’s capability to high-value clients on first view, without requiring any explanation

Insights

The Problem:


The MVP had no mechanism to surface OLarry’s core AI value. Insights existed in the system but had no place to live and no way to drive action.

The Solution:

Categorized insights feed: Organized by type, including savings, risk flags, wealth opportunities, and overspending, with estimated dollar impact on every card

Conversational AI threading: Each insight opens a direct AI conversation that explains the opportunity in plain language and provides actionable next steps

Intelligent escalation: Threads escalate to a human advisor when complexity warrants it, creating a natural handoff between AI and professional guidance

The consumer platform also includes Activities, Meetings, Documents, Filed Returns, My Tax Profile, and Settings; each designed with the same clarity-first principle.

step 7: designing the tax professional experience

the tax professional experience

Designed for experts managing an entire practice

Designed for experts managing an entire practice

The tax professional dashboard

The Problem:


No professional view: Tax advisors could only see a flat list of their clients, with no context, status, or filing information attached

No season-level awareness: There was no mechanism to surface where clients stood in the filing process or flag who was falling behind

No risk visibility: Advisors had no way to identify which clients were at risk across their entire portfolio at a glance

The Solution:

Client-centered dashboard architecture: Organized the dashboard around the four questions every client asks at the start of tax season, mapping each directly to a KPI card with year-over-year trend context

Prioritized action feed: Surfaced exactly what each client needs to act on next, with time estimates per task to reduce friction and set clear expectations

AI insights panel: Designed a dedicated panel that surfaces savings opportunities, risk flags, and wealth moves with estimated dollar impact per recommendation

Advisor conversion mechanism: Positioned the insights panel as the core business driver, translating client engagement into scheduled advisor actions

Activities

The Problem:


Tax professionals managing large client books had no way to understand what needed their attention across all clients simultaneously. Work was reactive — advisors had to manually check each client record to find blockers, unanswered messages, or stalled returns.

The Solution:

Prioritized action feed: shows the most urgent items across the entire client book: stale reviews, kicked-back returns, unanswered client questions. All ordered by urgency so advisors know exactly where to start

Split-panel detail view: selecting any action opens the full context on the right without leaving the screen, keeping advisors in flow rather than navigating between records

Gather + Review Workflow

The Problem:


Document intake was entirely manual. Advisors opened, read, and categorized every uploaded file before beginning a return, a significant source of time loss that scaled badly across a large client book.

The Solution:


Three-panel Gather workflow: Document request list on the left tracking received vs. missing, document viewer in the center, and AI-extracted field data with confidence scores on the right Verify, not transcribe: Advisors review and correct AI-extracted data rather than manually entering it, removing the most time-consuming part of document processing

Directly connected to Review: a worksheet-based reconciliation surface with projection vs. actual columns, per-schedule drill-down, and inline flags, the full picture from intake to sign-off in one continuous flow

Contextual AI bar: Allows advisors to ask questions about any document in the context of that client’s full tax picture, surfacing relevance without switching screens

step 7: implementation

Design Handoff & Implementation

After completing the design solutions, I facilitated a comprehensive design review and handoff process to ensure successful implementation:

1

Built an annotation and tracking system in Figma documenting interactions, behaviors, edge cases, and flows directly on each frame

2

Created a design status system (Ready for dev, In progress, Needs review, Dev complete) so the team always knew the state of each component, screen, or flow

3

Established a weekly design/engineering sync to compare live builds against Figma frames, flag regressions, and create tickets for fixes

4

Presented each solution with supporting rationale, gathered technical feasibility feedback, and iterated before handoff

Step 8: learnings and next steps

Current Status and Next Steps

This project represents an ongoing iteration cycle focused on continuous user validation and improvement:

Cross-Functional Collaboration:
Partnering closely with engineering to ensure designs are feasible and handed off clearly.

Research-Driven Iteration:
Interviews with tax professionals and consumers will directly inform the next round of changes

Continued Design Refinement:
Ongoing polish to sharpen both the UX and the premium feel of the product

Upcoming Validation Phase

1

Stakeholder presentation: Presenting current designs to gather feedback and align on direction before moving into the next round of changes

2

User interviews: Conducting video calls with both tax professionals and consumers to better understand what isn’t working and what needs to improve

3

UI polish: Continuing to refine the visual design to sharpen the premium feel of the product

4

UX iteration: Ongoing improvements to the experience for both sides of the platform based on research findings

Key Takeaways

Defining Done: QA and Remote Collaboration

Design doesn't end at handoff. Building a QA review step into the process, conducted live with engineering on a call, was essential to shipping a product that was both visually precise and functionally correct.

Knowing When an Edge Case Is Worth Solving

Managing a dual-sided platform includes a lot of scenarios. Learning to distinguish between a common use case and an edge case, and advocating for validation before letting either one drive the roadmap, became a critical part of the process.

Designing for Where the Product Is Going

Every design decision was made with the future in mind. Building scalable components and patterns from the start means the product can evolve without requiring a full redesign every time something changes.

step 8: implementation

Design Handoff & Implementation

After completing the design solutions, I facilitated a comprehensive design review and handoff process to ensure successful implementation:

1

Built an annotation and tracking system in Figma documenting interactions, behaviors, edge cases, and flows directly on each frame

2

Created a design status system (Ready for dev, In progress, Needs review, Dev complete) so the team always knew the state of each component, screen, or flow

3

Established a weekly design/engineering sync to compare live builds against Figma frames, flag regressions, and create tickets for fixes

4

Presented each solution with supporting rationale, gathered technical feasibility feedback, and iterated before handoff

Step 8: learnings and next steps

Current Status and Next Steps

This project represents an ongoing iteration cycle focused on continuous user validation and improvement:

Cross-Functional Collaboration:
Partnering closely with engineering to ensure designs are feasible and handed off clearly.

Research-Driven Iteration:
Interviews with tax professionals and consumers will directly inform the next round of changes

Continued Design Refinement:
Ongoing polish to sharpen both the UX and the premium feel of the product

Upcoming Validation Phase

1

Stakeholder presentation: Presenting current designs to gather feedback and align on direction before moving into the next round of changes

2

User interviews: Conducting video calls with both tax professionals and consumers to better understand what isn’t working and what needs to improve

3

UI polish: Continuing to refine the visual design to sharpen the premium feel of the product

4

UX iteration: Ongoing improvements to the experience for both sides of the platform based on research findings

Key Takeaways

Defining Done: QA and Remote Collaboration

Design doesn't end at handoff. Building a QA review step into the process, conducted live with engineering on a call, was essential to shipping a product that was both visually precise and functionally correct.

Knowing When an Edge Case Is Worth Solving

Managing a dual-sided platform includes a lot of scenarios. Learning to distinguish between a common use case and an edge case, and advocating for validation before letting either one drive the roadmap, became a critical part of the process.

Designing for Where the Product Is Going

Every design decision was made with the future in mind. Building scalable components and patterns from the start means the product can evolve without requiring a full redesign every time something changes.