<aside>

The Challenge

Document management B2B2C SaaS for the financial industry had no analytics and no clear differentiator against Google Drive or Dropbox.
Financial advisors spent 1-2 hours/day manually scanning and filing documents into client folders.
The mandate: find the growth unlock to justify the next round of funding.

Key problems

No user analytics to guide product decisions.
No answer to "why us vs. Google Drive?"
Advisors buried in manual filing instead of serving clients. </aside>

<aside>

What I Did

Built a data-first roadmap priority system focused on post-signup success.
Integrated ML-powered document auto-classification into the onboarding flow, reducing friction and accelerating time-to-value.
Pushed for and shipped ML auto-filing - the feature that became the company's differentiator.
Led a 20+ person cross-functional team (engineers, designers, DevOps, product marketing, customer success).
Launched user research with financial advisors to mine insight, not opinion.
Tracked leading indicators (ML confidence scores, upload volume, unfiled folder size) to prove the model was learning before adoption numbers showed it.

How I Think

In a regulated industry, a misfiled document is a legal problem. An unfiled document is just inconvenient.
I built a 95% confidence threshold into the ML model so we optimized for precision over recall.
My rule: "What's our fallback if this is wrong?" before any feature goes to pilot. </aside>

<aside>

AI/ML Architecture I built there

---
config:
  layout: elk
  elk:
    nodePlacementStrategy: NETWORK_SIMPLEX
---
flowchart TD
    subgraph Input ["<b>Document Sources</b>"]
        B1["Web Upload"]
        B2["Mobile App Scan"]
        B3["Physical Document Scan"]
        B4["Email Forwarding"]
    end

    B1 & B2 & B3 & B4 --> C["Document Ingestion Pipeline"]

    C --> D["OCR Engine"]
    D --> E["ML Classification Model"]

    E --> F{"Confidence Score"}
    F -->|"≥ 95%"| G["Auto-Filed to <br> correct folder + client"]
    F -->|"< 95%"| H["Unfiled Queue for human review <br> <i>(Human-In-The-Loop)</i>"]

    G --> I["Client Vault"]
    H --> J["Advisor Reviews and Files Manually"]
    J --> I

    subgraph Access ["<b>B2B2C Access Layer</b>"]
        K1["Financial Advisor View"]
        K2["Client View"]
        K3["Enterprise Admin View"]
    end

    I --> Access

    style Input fill:#F0F4F8,stroke:#37474F,stroke-width:2px,color:#333
    style C fill:#F0F4F8,stroke:#37474F,stroke-width:2px,color:#333
    style D fill:#F0F4F8,stroke:#37474F,stroke-width:2px,color:#333
    style E fill:#E8DFF5,stroke:#5E35B1,stroke-width:2px,color:#333
    style F fill:#FCE4EC,stroke:#AD1457,stroke-width:2px,color:#333
    style G fill:#E8F5E9,stroke:#2E7D32,stroke-width:2px,color:#333
    style H fill:#FFF8E1,stroke:#F57F17,stroke-width:2px,color:#333
    style I fill:#EDE7F6,stroke:#5E35B1,stroke-width:2px,color:#333
    style J fill:#FFF8E1,stroke:#F57F17,stroke-width:2px,color:#333
    style Access fill:#EDE7F6,stroke:#5E35B1,stroke-width:2px,color:#333

</aside>

<aside>