Spokane, WA / Building in public safety and applied AI

Dan Deppert

Turning What’s Next into What Works

I build AI-assisted hazard intelligence for first responders, run enterprise security accounts, and ship AI systems that people actually use. Thirteen years in enterprise technology, twenty in getting things done.

HazNetix Co-Founder & CTO
AlgoSec Technical CSM
Aadi Space Head of Sales

See HazNetix See the code

intake.py

from datetime import datetime
from fastapi import FastAPI
from pydantic import BaseModel, Field

app = FastAPI()

class HealthRecord(BaseModel):
    model_config = {"extra": "forbid"}

    account_id: str
    recorded_at: datetime
    health_score: float = Field(ge=0.0, le=1.0)

@app.post("/health")
def ingest(record: HealthRecord) -> Assessment:
    # Bad input fails loudly, never silently
    if record.health_score < AT_RISK:
        return flag_for_outreach(record)
    return assess(record)

Now

Three jobs, one thread

All three come down to the same skill: take something messy and unproven, and turn it into a system somebody can run.

Primary build Founded 2025

HazNetix

Co-Founder & Chief Technology Officer

HazNetix is building an AI-assisted hazard intelligence and response-orchestration platform, starting with AI-assisted 360 size-up for first responders. It is software-first, hardware-agnostic, and designed to preserve human command authority. I lead product and technical execution.

Operator console and the AI-assisted workflows behind it
System architecture and proof-of-concept planning
Design-partner validation with operators and domain experts

See how it works

Enterprise 2026 – present

AlgoSec

Technical Customer Success Manager

Enterprise security and network accounts across financial services, energy, government, transportation, and manufacturing. The work sits where security posture, network change, and customer outcomes meet.

Owns retention and expansion across the portfolio
Executive business reviews on a value-realization and ROI framework
Built a Python PPTX workflow that generates the review decks

Advisory Contract

Aadi Space

Head of Sales

Building the full sales and customer success motion from scratch for Orchestrator, their ground-station management platform. Challenger method, teaching and tailoring rather than pitching.

70+ company tiered prospect list built from zero
Three-touch outreach cadence, HubSpot as the system of record
Tier 1 segments: test & validation providers and universities

HazNetix

AI does the sorting. The operator still commands.

An AI-assisted hazard intelligence and response-orchestration platform. We start with 360 size-up for first responders: taking fragmented incident information and organizing it into recommendations a person can review, approve, and answer for later. Software-first and hardware-agnostic. Click through the workflow.

Organize

Incident information arrives fragmented, from sources that were never built to talk to each other. It gets organized into one shared picture before anyone is asked to act on it.

One organized view instead of scattered feeds
Partial or conflicting inputs are flagged, not quietly guessed at
Hardware-agnostic, so no single source is a dependency

Gate Nothing moves forward unorganized.

The team behind the build is mostly agents

HazNetix runs on a virtual team of ten specialist agents and twenty-three modular skills that I designed and built. It covers frontend through deployment, plus legal, product, and go-to-market. Safety is a first-class role in that team, not an afterthought: model evaluation acts as a gatekeeper, explainability is built for incident commanders, and safety-critical code has a mandatory review path.

Specialist agents: 10
Modular skills: 23
Human decision gates: Every step

Under the Hood

I write the code, not just the deck

Most people who do my job stop at the slide. Below is a working schema contract you can break yourself, and real code from three things I have built.

Break my schema contract

Contract-first intake is how I build any system that takes data from somewhere I do not control. This one scores account health. Edit the payload or pick a preset. It runs in your browser and answers three ways: reject bad input, flag what needs a human, pass what is clean.

Incoming payload · JSON

Contract: account_id str, tier one of strategic / enterprise / mid_market / smb, recorded_at ISO-8601, health_score float 0.0–1.0, notes optional. Unknown fields are forbidden.

Validator output

A browser reimplementation of a pattern I use, not a live system. The real version is Python and Pydantic on FastAPI. Same rules, same three outcomes.

Part of a churn-risk framework I built. Weighted signals, an explicit reason behind every score, and no silent defaults when a signal is missing. A health score nobody can explain is a number nobody trusts.

WEIGHTS = {"usage": 0.4, "support": 0.25, "engagement": 0.2, "sentiment": 0.15}

def score_account(signals: dict[str, float], tier: str) -> Assessment:
    """Blend weighted signals into one score plus the reasons behind it."""
    missing = [key for key in WEIGHTS if key not in signals]
    if missing:
        # Never quietly treat an absent signal as a zero
        raise IncompleteSignals(missing)

    score = sum(signals[key] * weight for key, weight in WEIGHTS.items())
    drivers = sorted(
        ((key, signals[key] * weight) for key, weight in WEIGHTS.items()),
        key=lambda pair: pair[1],
    )

    return Assessment(
        score=round(score, 3),
        band=band_for(score, tier),
        weakest=drivers[:2],
        needs_outreach=score < AT_RISK[tier],
    )

One of the automations behind the 86% case-volume reduction. It pulled open cases, flagged the stalled ones, and mailed a daily digest so nothing aged out silently.

function Get-CaseHealth {
    [CmdletBinding()]
    param(
        [Parameter(Mandatory)][string]$AccountId,
        [int]$StaleAfterDays = 5
    )

    # One timestamp for the whole run, so a slow page of results
    # cannot shift the age calculation mid-loop
    $now   = Get-Date
    $cases = Invoke-RestMethod -Uri "$ApiRoot/cases?account=$AccountId" -Headers $AuthHeader

    $open = $cases | Where-Object Status -ne 'Closed' | ForEach-Object {
        $age = ($now - [datetime]$_.LastUpdate).Days
        $_ | Add-Member -NotePropertyName AgeDays -NotePropertyValue $age -PassThru
    }

    $stale = $open | Where-Object AgeDays -ge $StaleAfterDays

    foreach ($case in $stale) {
        Write-Verbose "Stale $($case.AgeDays)d: $($case.Number) - $($case.Subject)"
    }

    [PSCustomObject]@{
        Account   = $AccountId
        OpenCases = $open.Count
        Stale     = $stale.Count
        OldestDay = ($stale | Measure-Object -Property AgeDays -Maximum).Maximum
    }
}

Executive business reviews used to take a day each. This builds the deck from structured account data on a value-realization framework, so the time goes into the story instead of the formatting.

def build_ebr(account: Account, period: str) -> Path:
    """Generate the value-realization deck for one account."""
    deck = Presentation(TEMPLATE)

    add_title(deck, f"{account.name} - Executive Business Review", period)
    add_value_summary(deck, account.outcomes, account.baseline)

    for theme in account.themes:
        slide = add_slide(deck, layout="theme")
        write_metrics(slide, theme.metrics)
        write_narrative(slide, theme.summary, theme.next_step)

    add_roadmap(deck, account.open_risks, account.expansion_paths)

    out = OUTPUT_DIR / f"{account.slug}_{period}_EBR.pptx"
    deck.save(out)
    return out

What I Build

Sixty-plus agents, skills, and GPTs in production use

I never finished a degree. I finished things instead. Everything below is something I designed, built, and put in front of real users.

60+ Agents & skills built

8 Custom GPTs shipped

3 Organizations using them

Active build

HazNetix Operator Console

A simulated operator console and validation platform, currently in design-partner testing with operators and domain experts. Software-first and hardware-agnostic, with approval workflows and audit history built into the product rather than bolted on.

Product & technical lead
Operator workflows
Design-partner validation

Complete

HazNetix Agent Team

Ten overarching agents and twenty-three modular skills covering full-stack development, AI and ML, legal operations, product, marketing, finance, operations, sales, and QA. Model evaluation and safety sits as gatekeeper across the whole team.

Multi-agent design
Safety gating
XAI
Audit modeling

In use

AlgoSec Agent Workspace

A five-agent workspace I built for my own book of business: Customer Coverage Guardrail, Customer Meeting Closer, Risk Radar, Customer Reply Drafter, and Chief of Staff. Plus a QBR/EBR Builder GPT and a Daily Focus Coach wired into Outlook and calendar.

ChatGPT Enterprise
Custom GPTs
Agent orchestration

In use

Python QBR Deck Pipeline

Executive business reviews used to eat a day each. This generates the PPTX from structured account data on a value-realization and ROI framework, so the time goes into the story instead of the formatting.

Python
python-pptx
Value realization

In use

Aadi Space AI Startup Team

A thirteen-skill suite that runs the go-to-market motion for a ground-station platform: prospect research, tiering, outreach sequencing, and CRM hygiene against HubSpot.

Claude Skills
HubSpot
Challenger method

Published

Packaged & Career Skills

Reusable skills built to be handed to other people, not just run by me. Includes an instructions builder, a voice-matching cleanup step other skills call, and a career suite covering positioning, documents, LinkedIn, technical docs, and ghostwriting.

Claude Skills
Prompt engineering
Reusable design

Shipped

Enterprise PowerShell Suite

Eight-plus automation tools deployed enterprise-wide across Microsoft and Citrix environments. Reporting, email processing, and case management. This is the work that cut 1,400+ engineering hours a year and dropped manual effort by half.

PowerShell
SQL
Salesforce API
Reporting

Built and shipped, now archived

FocusBot

AI task prioritization assistant on Streamlit with a FastAPI backend and GPT-4o. Processed 1,000+ tasks in beta and returned roughly a 25% productivity gain in the pilot. The demo is retired, but the build is real and the code is mine.

Python
GPT-4o
FastAPI
Streamlit

Historical

Career GPT Fleet

A ten-GPT orchestrated system plus a sixteen-role virtual executive team, built during 2025. Cut about 40% of my own working time. Worth keeping on the record because it is where the multi-agent pattern I use now got figured out.

Multi-agent orchestration
Custom GPTs
Botpress

The stack underneath

Build

Python Advanced
PowerShell Advanced
SQL Advanced
FastAPI & Pydantic
Bash, JSON, XML, Markdown
Git, GitHub, VS Code, Postman

Applied AI

Prompt & instruction engineering
Multi-agent orchestration
RAG concepts
Cost optimization
Claude Skills & custom GPTs
Azure OpenAI, Azure Speech, Botpress

Enterprise

Salesforce, ServiceNow, HubSpot
Gainsight, Jira, Confluence
Power BI, Tableau
Citrix Cloud, NetScaler, DaaS
Microsoft 365, Active Directory
Incident management & RCA

Track Record

Numbers that held up

Thirteen years in enterprise technology, inside a twenty-year working career. These are the results that came out of it.

$70M

Peak strategic portfolio

Across 15 Fortune 100 and 500 accounts, 2020 to 2024. $120M+ managed cumulatively over the full tenure.

98%

Contract renewal rate

Dollar-weighted, 97 to 98% sustained. Zero churn in the final two years.

167%

Flagship account expansion

Grew a global e-commerce account from 6,000 to 16,000 deployments over roughly seven years.

$15M+

Pre-sales deals supported

Five-plus enterprise deals. Every one closed where I ran the technical validation. Supported, not led.

Recognition

Customer Impact AwardCitrix, 2016
Customer Excellence AwardCitrix, 2017
Customer Advocacy AwardCitrix, 2017
Premium Services PilotSelected as proof of concept lead, 2017

How I got here

2026 – now
HazNetix, AlgoSec, Aadi Space
Co-founder and CTO leading product and technical execution. Technical CSM. Head of sales. Three roles, one build practice.
2025
Independent AI Consultant
Designed AI workflow systems on OpenAI, Claude, and Botpress. Delivered one-on-one AI training to 20+ customer success, TAM, and sales engineering professionals.
2016 – 2025
Cloud Software Group (Citrix + TIBCO)
Technical Account Manager to Lead TAM over eight and a half years. Strategic portfolio peaking at $70M ARR across 15 Fortune 100 and 500 accounts, with a global team of four.
2013 – 2016
Covestic, on contract to Microsoft
Led a 14-member operations team supporting Microsoft's global e-commerce and subscription platforms at 99.9% uptime.
2012 – 2013
P2 Solutions Group, on contract to Microsoft
Systems engineer running zero-downtime migrations across Lync, Windows Server, SQL, and network services for 30+ dedicated clients.
2003 – 2012
Before technology
Restaurant operations and retail electronics. Ran a kitchen, ran a P&L, and carried a $500K sales quota with ten people reporting to me before I ever touched enterprise software. It is where the customer instinct came from.

Writing

AI Learning Journal

A newsletter about building AI agents, one challenge at a time. I write about what broke, what I tried, and what finally worked. No pitch, no hype, no pretending the first attempt went fine.

If you want the honest version of what building with AI is actually like, this is it.

Read the newsletter See the code

Free Resource

Take this and use it

Built from real deployments, not from a template. No email required.

AI & SaaS Implementation Checklist

A 30-60-90 day plan, eight pages

Pre-launch through scale, with sign-off gates, security validation, and timelines that match how implementations actually go. Written for customer success managers, technical account managers, and implementation leads who are tired of advice that ignores reality.

Pre-launch foundation. Security, data governance, and metrics settled before day one.
Phase 1 through 3. Pilot to steady-state operations, with a clear gate between each.
Honest timelines. 30-60-90 for pilots, 120-180 for enterprise. Context decides.

Use it. Adapt it. Just stop launching without foundations.

Download the PDF Get the editable DOCX

Contact

Let’s turn your what’s next into what works

Building something in public safety, hazard intelligence, or applied AI? Running an enterprise account team that needs a system? Want to talk about HazNetix? Any of those, I am glad to hear from you.

Email dan@dandeppert.com Book time 30-minute call LinkedIn in/dandeppert GitHub Dandeppert

Dan Deppert

HazNetix

AlgoSec

Aadi Space

Organize

Recommend

Approve

Record

The team behind the build is mostly agents

Break my schema contract

HazNetix Operator Console

HazNetix Agent Team

AlgoSec Agent Workspace

Python QBR Deck Pipeline

Aadi Space AI Startup Team

Packaged & Career Skills

Enterprise PowerShell Suite

FocusBot

Career GPT Fleet

The stack underneath

Build

Applied AI

Enterprise

Recognition

How I got here

AI Learning Journal

AI & SaaS Implementation Checklist

Let’s turn your what’s next into what works