Skip to content
The Newby Ventures logo and wordmark, reversed white and green on black
Hunter Newby speaking at a podium in front of a Newby Ventures branded backdrop at the W Atlanta

Research platform

An example page from the NV Research dashboard, designed and built end to end, including the data-API integrations and the interactive maps. NV Research brings complex network-infrastructure data to a broad audience through ease of use, interactivity, and considered design.
The NV Research dashboard: a United States Internet Exchange data page with stat tiles, a growth chart, a state choropleth map, data tables, and an interactive facility map

AI-assisted content creation and publishing

Hunter Newby is an active speaker on the industry conference circuit, and every appearance needs a web page, a promotional graphic, social posts, and follow-up. I built an agent-run pipeline that does all of it from a short brief: it composes the graphic against a fixed template that folds in each event’s branding while keeping Newby Ventures front and center, publishes to the CMS through a custom API, drafts and posts to LinkedIn and X, and files the follow-up tasks in our client dashboard.

The part that took the most thought was not the automation. It was deciding what the automation is not allowed to do on its own. Every asset passes a two-tier review before it reaches a person, and nothing publishes without human approval.

Four event-promotion social templates for DataCloud USA, the AI Interconnection Summit, DCD Broadcast, and Capacity Metro Connect
A Newby Ventures LinkedIn post promoting Hunter Newby speaking at Bisnow's Data Center Capital Markets Summit
A finished event graphic. The brief is four lines of text: event, dates, panel title, panelists. Everything else is composed automatically against the template.
A Newby Ventures event promotion graphic for Mountain Connect 2025, showing the Newby Ventures logo, the panel title Accelerating AI into Broadband Networks, Hunter Newby's headshot, and the dates August 4 to 6 in Denver.
Review runs in two tiers. Geometry, character counts, string integrity and link status are deterministic and stay in Python. Composition, legibility and factual accuracy need judgment and go to the model. Both resolve to a single decision, and neither one approves anything.
BRAND CHECK: PASS   (mountain-connect-2025, attempt 1)

Tier 1 — deterministic
[  ok  ] D1   Widest title line reaches x=511; event logo column ends at x=426. Gap 85px, clear.
[  ok  ] D2   Title block plus panel time ends at y=384; limit is y=515 (3 lines).
[  ok  ] D3   Closest title line sits 24.0px from the headshot circle (minimum 20px).
[  ok  ] D11  Panel title wraps to 3 lines (comfortable maximum 4). Geometrically legal.
[  ok  ] D6   No banned strings.
[  ok  ] D8   https://mountainconnect.org returned 200.

Tier 2 — judgment
[  2/2 ] J1   Balanced composition; the three-line title sits comfortably in its column with hierarchy intact.
[  2/2 ] J2   Mountain Connect logo reads clearly against the light panel.
[  2/2 ] J3   Dates (Aug 4-6, 2025), Denver CO, and panel time (Aug 4 @ 3:00 PM) match the brief.
[  2/2 ] J4   Panel title verbatim (template sets all-caps); no truncation or rewording.
[  2/2 ] J5   Generated text is concise, factual, peer-to-peer.
The graphic generator composes each asset from a fixed template, so the layout rules are known quantities rather than guesses. They are measured off the template once and enforced in code. Type clearance around the headshot is checked geometrically, not by asking a model whether it looks crowded.
HEADSHOT_LEFT = 958
HEADSHOT_TOP = 167
HEADSHOT_BOTTOM = 368
HEADSHOT_RADIUS = (HEADSHOT_BOTTOM - HEADSHOT_TOP) / 2
HEADSHOT_CX = HEADSHOT_LEFT + HEADSHOT_RADIUS
HEADSHOT_CY = HEADSHOT_TOP + HEADSHOT_RADIUS

def rect_circle_clearance(left, top, right, bottom, cx, cy, r):
    """Shortest distance from a rect to a circle's edge. Negative means overlap."""
    nearest_x = max(left, min(cx, right))
    nearest_y = max(top, min(cy, bottom))
    return math.hypot(cx - nearest_x, cy - nearest_y) - r
Deterministic checks and model judgment resolve to one verdict. Warnings never block. Failures send the asset back once, then stop and ask a person.
def route(tier1, tier2):
    if any(c["result"] == "fail" for c in tier1):
        return "flag"
    if tier2:
        if any(c["score"] == 0 for c in tier2):
            return "flag"
        if sum(c["score"] for c in tier2) < 8:
            return "flag"
    return "pass"

The point of the system is not to take people out of the work, but to spend their attention where it counts. The pipeline carries the repetitive production so the team’s judgment goes to what actually needs it: whether the composition holds, whether the facts are right, whether it is good enough to carry the client’s name. That is what lets a small team keep a large event calendar moving without letting brand quality slip.

Custom AI assistant

A custom assistant trained only on Hunter's own material: his articles, the AI Interconnection book, and the PeeringDB-driven research pages. Visitors can ask about the book, the research tools, or the industry topics he covers and get answers grounded in his work rather than the open web.
The AI Hunter Newby assistant open over the NV Research dashboard, answering a visitor's question about why latency matters for AI with a detailed response drawn from the research

Brand & marketing

A cover for a digital-only book. With no physical edition to print, the design lets the cover dissolve into the page at its base, a nod to a book that lives entirely on screen.
Landing page for the AI Interconnection book, its 3D cover dissolving into pixels at the base
Templated email newsletters and lifecycle automation nurture the audience over time: a monthly interconnection-data roundup and research-series announcements that keep subscribers engaged and draw them back to the tools.
Two Newby Ventures email newsletters: a research-updates letter and a monthly interconnection data roundup
Hunter Newby has published 130+ articles and spoken at 150+ industry conferences. To catalog his standing as a thought leader, we compiled his full body of work and appearances into a filterable timeline on the site.
Hunter's Media & Appearances: a filterable timeline cataloging his articles, videos, podcasts, and conference appearances