Logo

Senior Program Manager · KnowBe4

Cole S. Kujawa

Program leadership with an engineer's toolkit.

Eight years across customer success, product, and cross-functional delivery. I turn ambiguous stakeholder input into specs engineers can build from, keep delivery data trustworthy, and prototype with AI tools. In my own time I build, measure, and publish the kind of systems I manage.

PDF copy last updated Oct 2026 · v2.4

8+years in tech, from support to program leadership
6industry credentials
7featured projects below
30+production automations running

Credentials

Certified by the platforms I build on

Hover or tap a badge for details. Where a public verification page exists, it is linked on the back.

  • Claude Certified Architect - Foundations badge (Anthropic)Claude Certified Architect - FoundationsAnthropic

    Anthropic

    Earned 2026

    Proctored exam on building production-grade apps with Claude and MCP.

    Verify credential
  • AWS Certified Cloud Practitioner badgeAWS Certified Cloud PractitionerAWS

    AWS

    Earned 2022

    Foundational AWS knowledge: core cloud concepts, key services, security, and pricing.

    Verify credential
  • CompTIA A+ certification badgeCompTIA A+CompTIA

    CompTIA

    Earned 2020

    Entry-level IT support credential: hardware, operating systems, networking, and troubleshooting.

    Verify credential
  • CompTIA ITF+ certification badgeCompTIA ITF+CompTIA

    CompTIA

    Earned 2020

    IT Fundamentals: core IT concepts, infrastructure, software, and security basics.

    Verify credential
  • Maven logoPrompt Engineering for LLMsMaven

    Maven

    Earned 2023

    Maven course on designing, testing and iterating prompts to get reliable results from large language models.

  • AIPMM Agile Certified Product Manager / Product Owner badgeAgile Certified Product Owner / Product ManagerAIPMM

    AIPMM

    Agile product ownership and management: aligning product strategy, backlogs, and delivery.

Toolbox

What I work with, and how deeply

Everything here is something I have either shipped with or studied properly.

Languages

Daily drivers first, coursework last.

Frameworks & platforms

What this site, and most of my services, run on.

AI & data

Applied, local-first, and measured.

Automation & operations

The part that keeps working when I am not looking.

Projects

Things I have built and published

Side projects, treated like real work: measured, documented, and rerunnable where possible.

Spotted Lanternfly Detection-Gap Model interactive map

2026 · Geospatial machine learningShipped

Spotted Lanternfly Detection-Gap Model

An invasive insect has not been found near Great Smoky Mountains National Park, but is that because it is absent, or because nobody is looking? I built a habitat model and an observer-effort surface, then combined them into a map of where surveys would pay off most.

  • Cleaned and spatially thinned 32,641 citizen-science records to 8,835 presences; none fall within 50 km of the park.
  • Stacked six environmental layers (climate normals, elevation, land cover) on a common 250 m equal-area grid.
  • Trained a 500-tree Random Forest and validated it with spatial block cross-validation (AUC 0.75 ± 0.13), because random splits flatter geospatial models.
  • Published the write-up and a five-notebook, rerunnable pipeline, including the limitations I would fix next.
Pythonscikit-learnGeoPandasFoliumJupyter
01Voice or chat request
02Sentence triggers · under 100 ms
03Intent scripts and slash-command filters
04Local LLM with tools (Ollama)
05Spoken summary + full chat detail

2025–now · Applied AI & automationOngoing

Jarvis: a local-first AI assistant

A voice-and-chat assistant that runs entirely on my own hardware. Home Assistant, Ollama, Open WebUI and n8n are wired so known requests are handled deterministically in milliseconds, and the model only sees what truly needs it.

  • Layered routing: sentence triggers answer in under 100 ms; anything unmatched falls through intent scripts, then to the model.
  • Slash-command filters write their own answers, so correctness never depends on the LLM. After one leaked a fabricated reply, I added an audit that checks every filter.
  • One tool response serves two surfaces: a single spoken line for voice, full markdown for chat.
  • 30+ n8n workflows, a five-leg smoke-test suite, and 22 architecture decision records behind it.
Home AssistantOllamaOpen WebUIn8nClaude CodeDocker
YOU80 km8 km240°–300°

2026 · Streaming data & alertingShipped

Thunderhead: real-time storm alerting

A personal early-warning system that watches a slice of the Gulf for forming thunderstorms and pushes a phone alert when the numbers say it is worth looking up.

  • Subscribes to a live lightning-strike feed over MQTT for 12 geohash tiles, filtered to an 8–80 km sector and bucketed into roughly 5 km grid cells.
  • Fires on a 2σ jump in strike rate, a new cell with neighbor checks, or a density threshold; 45-minute cooldowns prevent alert spam.
  • Confirms against airport weather reports, and closes the loop: thumbs-up or thumbs-down feedback is written back to Postgres to tune thresholds.
  • Cross-checked against GOES-19 satellite lightning data, with a liveness monitor that stays quiet on genuinely quiet days.
n8nMQTTPostgreSQLHome AssistantPython
01Question
02Embed (nomic-embed-text)
03Vector search (Qdrant)
04Open the live source file
05Grounded answer + citation

2026 · Retrieval-augmented generationOngoing

Knowledge-base search (RAG)

Natural-language search over my architecture decision records, incident write-ups and plans. It finds the right document, then answers from the live file with a citation instead of a stale chunk.

  • Embeds with nomic-embed-text, retrieves from Qdrant, and grounds a local LLM on the current file, so answers cannot drift from an edit made after the last index.
  • Incremental ingestion by content hash, safe to run on every change.
  • A sibling endpoint answers questions about books with page-and-paragraph citations and verbatim-quote verification.
  • Reachable from chat, voice, or the command line, with all inference kept on the host.
QdrantOllamaPythonDocker

Interactive graphicsShipped

Real-time fluid simulation

The animation on my home page: a WebGL fluid simulation wrapped as a reusable Payload block, so I can drop it on any page and tune it from the admin. The live one is running right here.

  • Adapted an open-source WebGL solver into a configurable block: palettes, emission modes, bloom, and simulation resolution.
  • Handles WebGL context loss, reduced-motion preferences and touch devices, and is code-split so other pages never download it.
  • Collides with rendered text, so the name and subtitles on the home page sit inside the flow.
TypeScriptWebGLPayload CMSNext.js
Seven years of forest change, from space

2026 · Remote sensingShipped

Seven years of forest change, from space

Can free public satellite data show where a forest is gaining and losing greenness? Seven years of Sentinel-2 imagery over the Smokies say yes: about 6% of the study area changed meaningfully.

  • Pulled every July–August scene from 2018 to 2025 with under 20% cloud, streamed as cloud-optimized GeoTIFFs from the Planetary Computer.
  • Cloud-masked with the scene classification layer, median-composited per year, and computed NDVI change.
  • Published as notebooks plus a companion guide to using the Planetary Computer from a home lab.
PythonJupyterSentinel-2STAC
LiDAR terrain · GNSS traverse · volume check vs QGIS

2026 · Geospatial productIn progress

Field survey & site-grading web app

A field-ready mapping app for capturing a site with GNSS, comparing existing and proposed ground, and exporting data an engineer can actually open.

  • Offline-first: a service worker, a local mirror and a write outbox that syncs when signal returns.
  • GNSS capture tuned for survey receivers: accuracy classes, averaged occupations, and stale-fix warnings.
  • Existing-versus-design grading over LiDAR terrain, exporting to GeoJSON, State Plane coordinates and a ready-to-open QGIS project; volumes agree with QGIS to within 1.3%.
TypeScriptLeafletPayload CMSMongoDBQGIS

Experience

Where I have done the work

  1. Senior Program Manager · KnowBe4

    Sep 2026 – Present · Clearwater (Remote), FL
    • Own Jira as the system of record for delivery: project structure, workflows, permissions, automation, and reporting standards.
    • Set and enforce delivery hygiene (ticket quality, estimation, status accuracy) so leadership can trust delivery data without manual reconciliation.
    • Produce delivery reporting and roadmap views for engineering leadership and stakeholders.
    • Primary connection between engineering and product, design, data, and GTM; surface cross-team dependencies, risks, and blockers early enough to protect commitments.
    • Partner with engineers to prototype and validate ideas quickly with AI tools, and turn ambiguous stakeholder input into structured specs and briefs.
    JiraDelivery reportingRoadmapsRisk & dependenciesAI-assisted prototyping
  2. Technical Program Manager · KnowBe4

    Jul 2022 – Sep 2026 · Clearwater (Remote), FL
    • Led multiple cross-functional teams (up to 15 people) through the full program lifecycle, from requirements gathering to deployment.
    • Ran concurrent projects to on-time delivery and budget with detailed plans, milestones, and resource allocation.
    • Coordinated engineering, product, automated test, and QA teams to define scope, timelines, and deliverables; brought in third-party vendors where needed.
    • Introduced Agile practices and reported status to senior leadership through Jira and Trello.
    • Authored requirements specifications, project plans, and status reports.
    AgileCross-functional leadershipVendor coordinationJiraTrello
  3. Product Coordinator · LightSpeed Voice

    Dec 2019 – Jul 2022 · Nokomis, FL
    • Carried product strategy and roadmap from requirements through to sunsetting features, including documentation and training.
    • Managed feature requests and bugs in GitHub; prioritized and planned sprints and releases across the portfolio.
    • Planned and documented APIs with third-party companies and supported their integrations, including demos and capability-gap analysis.
    • Ran User Group communications and meetings.
    RoadmappingAPI documentationPartner integrationsGitHubRelease planning
  4. Client Success Team Lead · LightSpeed Voice

    Mar 2019 – Dec 2019 · Nokomis, FL
    • Led the technical support team through widespread customer issues while keeping ticket and call volume low.
    • Built methods for tracking call and ticket statistics, feature requests, and development progress.
    • Hosted user panels to build rapport and surface features; represented support in developer meetings across teams.
    Team leadershipSupport operationsMetricsCustomer panels

Education

Foundations

B.S., Information Technology

University of Phoenix

A.S., Computer Science (Software Development emphasis)

Palomar College

Relevant coursework

Programming Fundamentals I & IIData Structures (in Java)Machine Organization / Assembly LanguageC++ & OOPJava Programming IAndroid DevelopmentWeb Design FundamentalsDatabase ConceptsApplication ImplementationNetwork Web ServicesFundamentals of NetworkingNetwork Design AssociateGlobal Network Architecture & DesignUnix NetworkingAdministering Windows ServersWindows Server ConfigurationCloud TechnologiesProject Planning & Implementation