Start Here
Angel Roman, founder of Applied Tech Labs

Forward Deployed Engineer

INDEPENDENT AI
IMPLEMENTATION PARTNER

I’m a hands-on partner building AI operating systems that reduce operating costs and fulfill more customer demand with the same headcount. I own delivery from design and development through integration and team adoption. Everything is built around your existing software and equipment to minimize disruption.

The success of every engagement is measured by one clear business outcome and how much your team actually uses what I build.

  • Live within 90 days
  • no unnecessary migrations
  • enterprise-grade security
  • outcome-driven builds
Work with me ↗ See my prior work ↗

CORE PHILOSOPHY

Don’t just automate tasks
Transform the entire process

Often times, an automation or agent becomes another forgotten tool handling just one step in a larger process (and sometimes, that step shouldn’t even exist). You don’t need more AI tools. You need one connected system that manages the entire process.

We start by identifying the core process the company or department depends on to fulfill demand. Then we redesign and connect the workflows, equipment, and expertise behind it into one AI-enabled system that manages it end-to-end.

The system handles the manual coordination and information flow, allowing people to apply their expertise without manually managing every step.

The result: an operating system that helps the entire department meet more demand and reduce backlog with the people and equipment you already have.

Client Results

Northbound & Co.

2 hours per client every week, reduced to 30 minutes
Google AdsMeta AdsGA4

NORTHBOUND & CO. · MARKETING REPORTING

Weekly client reporting cut from 2 hours to 30 minutes

“The reporting dashboard Angel built for Northbound & Co. has completely transformed my weekly reporting process. I manage reporting across multiple advertising platforms, and each platform exports its data in a completely different format. Previously, I had to manually organize, format, and combine everything before I could even begin reviewing performance.

The dashboard has made my reporting faster, more consistent, and much easier to manage across multiple accounts. Most importantly, I can spend more time analyzing performance and making strategic decisions instead of manually formatting spreadsheets every week.”

Scroll to read the full quote

Rubi Martinez, Founder, Northbound & Co.
4+ hours saved and multi-day delays removed in customer onboarding
Built within Google Workspace

DEEPLOOK MEDICAL · CUSTOMER OPERATIONS

4+ hours saved and multi-day delays removed in customer onboarding

Built a custom dashboard within DeepLook’s existing Google Workspace to manage partner referrals, clinical onboarding, activation, and adoption. Automated intake, pipeline progression, customer emails, and training setup saved 4+ hours and removed multi-day delays in customer onboarding, with human verification and interaction at critical steps.

100,000+ research records
Published research Clinical experience

SPINEDAO · AI RESEARCH PRODUCT

100,000+ research records. One new AI product.

Combined published spine research with proprietary clinical data from SpineDAO’s surgeon network to build an AI research assistant for their private journal. Readers could ask questions, explore complex findings, and connect published evidence with clinical experience, creating an entirely new product offering for the journal.

ASU Health Observatory

In progress

I’m building an interactive data explorer that brings together government records, university research, and partner contributions to help users uncover relationships, trends, and patterns across sources. An integrated AI chat assistant will answer questions using only this information and a curated set of high-quality references.

Nellis Auction

In progress

I’m consolidating more than 20 Data Studio dashboards and 50+ metrics into focused views for processing, logistics, warehouse, and operations managers. Building on Nellis Auction’s existing data infrastructure, each view will surface the few metrics that matter most and flag potential operational delays. The goal is to align managers around increasing daily product throughput and reduce the time from freight arrival to sale listing.

MY PROCESS

What working with me looks like

  1. STEP 01

    Initial qualifying call

    For a lot of companies, a custom data strategy and AI implementation is not the right investment. Our first conversation identifies what’s slowing your business down, whether technology can meaningfully help, and whether a custom build is justified.

    A few qualifying factors include:

    • High operating volume

      The company handles enough orders, quotes, shipments, service requests, or expensive projects that require substantial manual coordination.

    • Information spans multiple systems and teams

      Completing one customer order involves multiple people, software, and systems that disagree on current status and requires manual tracing.

    • Valuable company-specific data

      The business has order histories, production records, technical documents, or customer information that could improve decisions.

  2. STEP 02

    Non-disruptive workflow assessment

    Once we establish it's a good fit, I assess how work actually gets done through short interviews, job shadowing, and process mining where applicable (in-person if possible). This helps me map out what the ideal process should look like versus how the process actually looks currently.

    Understand how work really happens
    Team
    interviews
    On-site
    observation
    Process
    data

    Your workflow, mapped

    Request Work Delivery

    Pinpoint handoffs, delays, and rework.

  3. STEP 03

    Determine how work should operate with intelligence built in

    Once I understand the current process, I design the AI-enabled version and build the system that combines the two. This involves deciding where predictable automation fits, where AI needs to adapt, and where people need to stay involved.

    A general rule of thumb:

    • Rule-based automation

      The outcome and path to get there is the same every time with predictable rules.

    • AI Agents

      The outcome is the same but the path to get there varies.

    • Human judgement

      The outcome varies with shifting priorities and has irreversible consequences if unverified.

  4. STEP 04

    Build and integrate

    The system is built by connecting your ERP, tools, data, and operational logic. Checks, approvals, and oversight are built in, and we test against real operating scenarios before rollout.

  5. STEP 05

    Train and enable

    We train the team that owns the workflow and equip internal champions to run, monitor, and extend the system. Your people learn to use AI in their daily work and make data-driven decisions.

THE OUTCOME

No more conflicting information across teams and software. No more manual coordination delaying quoting, production, or delivery. No more tools your team struggles to adopt.

WHY ME

My Differentiating Principles

Three core principles guide every system I build to ensure provable ROI before anything is built.

One business outcome

Before anything is built, we define and agree on one measurable business outcome that matters most. That outcome sets the scope, guides every design decision, and defines how we measure success, ensuring that a real business outcome drives the partnership.

Adoption comes first

The best system is useless if nobody uses it. We build into the software and workflows your people are already familiar with, removing only the steps and tools that cause more chaos. We also train and enable your people to understand AI and make data-driven decisions.

Minimal by design

One clear view or no new interface at all. I bring automations, agents, and real-time data together, focusing only on the few metrics that drive impact on margin and revenue. Your team gets what it needs without juggling multiple noisy dashboards.

Book a Free Call

Or email angel@appliedtechlabs.com

DEEPLOOK MEDICAL · CUSTOMER OPERATIONS AUTOMATION

4+ hours saved and multi-day delays removed in customer onboarding

I built a custom customer operations dashboard within DeepLook Medical’s existing Google Workspace, connecting partner referrals, clinical onboarding, product activation, and adoption. The system automated routine handoffs and follow-up, saving 4+ hours and removing multi-day delays in customer onboarding while preserving human verification at critical stages.

DeepLook Medical customer operations dashboard showing clinical sites, onboarding status, partner referrals, and follow-ups, with identifying details redacted.
Customer operations overview
DeepLook Medical clinical site pipeline organized into acquisition, outreach, activation, and onboarding stages, with identifying details redacted.
Clinical site pipeline

The challenge

DeepLook’s small team was managing a growing volume of clinical customers through an onboarding process shaped by the requirements of a regulated industry. Most customers arrived through partner referrals, adding another layer of coordination between partner teams, clinical sites, and internal staff.

Their needs extended beyond a standard sales pipeline. Each customer required hours of manual work across outreach, implementation, and adoption. Staff repeatedly reconciled partner spreadsheets, tracked prerequisites, coordinated activation codes, and sent onboarding materials.

Manual handoffs and follow-up left customers waiting days at different stages, and the team had little capacity left to support adoption after activation.

The architecture

Customer operations dashboard. A custom HTML/CSS/JavaScript interface brought partner relationships, clinical-site records, contacts, and pipeline status into one working view. Google Sheets held the underlying records, while Google Apps Script handled the dashboard’s data access and workflow logic.

Criteria-based onboarding. Scripts checked defined onboarding criteria and advanced customers when requirements were satisfied. Routine transitions happened automatically, with human verification retained at necessary checkpoints, including product activation.

Activation and customer communication. The workflow coordinated the activation-code handoff and automated product-information emails, reducing the manual chasing and repeated communication previously required for each customer.

Adoption support. Automated training-link setup made team training part of the onboarding workflow, giving customers a defined next step after activation and reducing the coordination needed from DeepLook’s team.

Partner referral intake. A custom script function brought partner-submitted referrals into the customer pipeline automatically. This replaced recurring spreadsheet reconciliation and reduced the need for meetings just to establish which customers needed attention.

Software stack

ComponentRole
HTML, CSS, JavaScriptCustom customer operations dashboard
Google Apps ScriptWeb-app hosting, backend logic, and workflow automation
Google SheetsPartner, clinical-site, contact, and pipeline records
Google Workspace / DriveExisting working environment

The result

4+ hours saved and multi-day delays removed in customer onboarding.

SPINEDAO · AI RESEARCH PRODUCT

Turning complex spine research into an interactive reader experience

I worked with SpineDAO to build an AI assistant for their private spine journal that combined published research with proprietary clinical information collected across its spine surgeon network. The goal was to create a new interactive product for their research journal, giving readers a more accessible way to explore complex evidence alongside knowledge drawn from clinical practice.

Spinal journal interface with published spine research, article details, and the Ask Vincent AI research assistant, with account and contributor information redacted.
The research journal
AI research response with a direct answer, evidence base, and findings organized by evidence hierarchy.
Structured research responses

The product opportunity

Spine research is highly specialized and difficult to interpret. Understanding a finding requires more than reading its conclusion; it requires readers to consider the patient population, surgical approach, outcomes measured, limitations, and how those differ across different studies.

Meanwhile, SpineDAO’s surgeon network held valuable clinical information beyond what appeared in published literature. The opportunity was to bring these sources together in an assistant that helped readers explore questions across both published research and the network’s own clinical knowledge.

Understanding the user

I had direct conversations with the spine surgeons leading the business to help translate the idea into practical requirements: what information mattered, how findings needed to be organized, and what made an output useful for research review.

This shaped the workflow around the reader’s task and how the system retrieved, processed, and presented information.

The architecture

I built a workflow connecting research ingestion, evidence indexing, and task-specific LLMs to a conversational interface.

Research ingestion and processing. Publication APIs collected research papers for processing. LLM workflows extracted key findings and organized them into searchable evidence, making information reusable across questions without reprocessing every publication.

Combined knowledge retrieval. The retrieval workflow connected indexed publications with SpineDAO’s proprietary clinical information. A reader’s question guided the selection of relevant evidence across both sources, bringing published findings and network-contributed knowledge into the same research workflow.

Task-specific synthesis. Retrieved evidence passed through LLM workflows configured to organize and summarize the material into structured responses. Separating retrieval from synthesis grounded the response in the connected information and allowed each part of the workflow to be refined independently.

Conversational access. Readers interacted with the system through natural-language questions. The assistant brought relevant findings together with source references, providing an entry point into the evidence and a path back to the underlying material for closer review.

Software stack

  • Flask + Next.js: Reader interface for submitting questions and displaying structured, source-linked answers.
  • Node.js + Express: Backend API connecting the interface, data sources, and AI workflows.
  • PostgreSQL: Structured storage for clinical registry information and surgeon-contributed case discussions.
  • Elasticsearch: Search across indexed publications from PubMed and medRxiv.
  • Neo4j: Knowledge graph connecting conditions, treatments, outcomes, and supporting evidence.
  • Open-source models: A cost-effective method for interpreting research questions and synthesizing retrieved evidence into cited responses.

NORTHBOUND & CO. · MARKETING REPORTING AUTOMATION

Cutting weekly client reporting from 2 hours to 30 minutes

Built a reporting platform for Northbound & Co. that brings advertising and analytics data into one place, making trends and patterns easy to explore across all clients and channels. The platform supports the full weekly reporting workflow: aggregating data, visualizing performance, reviewing findings, developing strategy, and writing final client reports, with AI-assisted summaries to support analysis.

Northbound reporting platform showing a combined cross-client and cross-channel performance overview and AI-assisted summary, with client information and selected figures redacted.
All clients and channels in one view
Northbound reporting platform showing performance over time and channel insights, with selected information redacted.
Trends, patterns, and channel insights

The challenge

Northbound managed client reporting across multiple advertising platforms, each exporting data in different formats. Every week, the team had to aggregate data, create spreadsheets to visualize it, read through the results, develop strategy, and write the final client reports.

The complete workflow took roughly two hours per client every week. Data preparation was only one part of that time. As the number of accounts grew, the work repeated across clients and channels, making it harder to see shared trends and patterns and act on them.

The architecture

Multi-source data intake. An n8n workflow loaded each client’s reporting configuration and retrieved uploaded Google Ads, Meta Ads, and GA4 exports from designated Google Drive folders. Client-specific settings determined which sources to process and where to save the outputs.

Data standardization and validation. Converted platform-specific exports into a consistent reporting structure and calculated metrics such as spend, clicks, conversions, and cost per result. Checks flagged missing files, parsing failures, and discrepancies between Google Ads account totals and campaign-level data.

Client reporting and dashboards. The platform brings all clients and channels into a shared view, with weekly and monthly comparisons that make trends and patterns easy to spot. The team can explore combined performance or focus on an individual client, channel, or tactic to guide strategy. The reporting engine generates Google Sheets outputs containing normalized metrics, source-status checks, and management summaries.

AI-assisted analysis. An OpenAI model was used to generate concise client and agency summaries from the structured metrics, highlighting performance and items requiring attention. Prompts instructed the model to use the supplied data and avoid claiming trends without prior-period comparisons. A predefined summary served as a fallback if AI output was unavailable.

Reusable multi-client workflow. A client registry allowed the same reporting workflow to run across accounts without duplicating the build. Each run logged results and flagged clients requiring review, allowing processing to continue when an individual client encountered an issue.

Software stack

ComponentRole
n8nWorkflow orchestration, client configuration, and run tracking
Google DriveSource-file intake and report organization
Google SheetsStructured reporting outputs and management summaries
OpenAI Company AccountAI-assisted client and agency summaries
Google Apps ScriptWeb-app hosting and backend logic

The result

The full weekly reporting workflow reduced from roughly 2 hours to 30 minutes per client, saving 1.5 hours per client every week.

Client feedback

“The reporting dashboard Angel built for Northbound & Co. has completely transformed my weekly reporting process. I manage reporting across multiple advertising platforms, and each platform exports its data in a completely different format. Previously, I had to manually organize, format, and combine everything before I could even begin reviewing performance.

The dashboard has made my reporting faster, more consistent, and much easier to manage across multiple accounts. Most importantly, I can spend more time analyzing performance and making strategic decisions instead of manually formatting spreadsheets every week.”