PROJECT 01
Ad Sales / Marketing
Sponsor Checker
THE PROBLEM
Finding sponsor appearances means reviewing footage by hand or paying an outside service.
WHAT I BUILT
An offline, open-weight tool combining logo recognition, on-screen text reading, and speech transcription into timestamped brand hits.
IMPACT
~$100K/yr
Veritone / AI Media contract it will replace
Searchable
every brand occurrence, timestamped
Offline
runs locally on open-weight models
2026 AI Impact Report · Joseph Yaksich
01 / 33
PROJECT 02
Marketing / Broadcast
AutoDubber
THE PROBLEM
We reach 10+ markets in four languages, but feeds stay English. Paid dubbing runs from $1–5 to $200+ per hour.
WHAT I BUILT
A local pipeline on an IT-issued M4 Mac mini, using four commercially licensed open-weight models to transcribe, translate, and re-voice speakers.
IMPACT
≤ 0.61 sec
timing drift, vs. 5+ sec in standard dubbing
$0
per hour of dubbed video, fully local
5 languages
Spanish, English, Chinese, Japanese, Arabic
2026 AI Impact Report · Joseph Yaksich
02 / 33
PROJECT 03
All departments / Executive
CourtSight
THE PROBLEM
Our first FAST pipeline covered ~3% of viewership. The other 97% sat behind complex, manual reporting systems.
WHAT I BUILT
After Marketing asked for weekly reports for Jeff, I spent two months building pipelines for the other 97%, normalized behind one API.
IMPACT
3% → 100%
FAST viewership coverage
Days → secs
for cross-domain questions
6-figure+
estimated first-year impact on rights, ad and renewal talks
2026 AI Impact Report · Joseph Yaksich
03 / 33
PROJECT 04
Executive / Content
Promo Pulse
THE PROBLEM
Asana pitched Asana AI at $11K/yr; its test, automating promo review for Jeff, had no results after 3+ months.
WHAT I BUILT
Scott in Content brought it to me. After one discovery meeting I had a working version in ~45 minutes and the full tool within 24 hours.
IMPACT
$11K/yr
Asana AI proposal avoided
24 hours
to the full tool, vs. 3+ months with no result
Daily use
including by Jeff
2026 AI Impact Report · Joseph Yaksich
04 / 33
PROJECT 05
Marketing / Broadcast
Automated Ticker
THE PROBLEM
Tagboard said the custom ticker we needed was impossible; Ross XPression cost $20–50K/yr and still fell short.
WHAT I BUILT
A web-based ticker fed by live ACE match data, with an operator interface to review content, set frequency, and schedule promos.
IMPACT
On air daily
T2 ticker in production; TC in development
~$50K/yr
current ticker license to be replaced
Full control
staff own content and cadence, no code
2026 AI Impact Report · Joseph Yaksich
05 / 33
PROJECT 06
All departments
TennisBot
THE PROBLEM
Leaders needed answers from viewership data without waiting days for an analyst to write SQL.
WHAT I BUILT
A model-agnostic natural-language layer on CourtSight: ask in plain English, get charts, SQL, and full exportable reports.
IMPACT
Days → secs
from question to answer
No SQL
non-technical users query the data directly
Any model
model-agnostic: swap models without a rebuild
2026 AI Impact Report · Joseph Yaksich
06 / 33
PROJECT 07
Ad Sales
GAM Ad Inventory Pipeline
THE PROBLEM
Ad Sales reporting took about three weeks from data capture to ready-to-report.
WHAT I BUILT
After seeing the FAST work, Ad Sales asked for the same. Over two weeks I automated it end to end, trained the team, and handed it off.
IMPACT
Same day
reporting, down from about three weeks
20+ hrs/wk
redirected into revenue-generating selling
1.7B+
impressions tracked across TC, T2 and PBTV
2026 AI Impact Report · Joseph Yaksich
07 / 33
PROJECT 08
The Big T / Marketing
The Big T Website
THE PROBLEM
The Big T premiered in early January 2026 with no site; vendor bids were ~$50K and couldn’t make the date.
WHAT I BUILT
Over Christmas week I worked remotely with several teams to design, iterate, and ship the complete site for $0.
IMPACT
~$50K
vendor bid avoided
On time
live for the premiere
Live
today in tennis.com’s Watch navigation
2026 AI Impact Report · Joseph Yaksich
08 / 33
PROJECT 09
Marketing / Consumer
tennis.com How to Watch
THE PROBLEM
Marketing needed a consumer landing page. Outside vendors quoted $50K and a three-month timeline.
WHAT I BUILT
Built in-house in about two weeks, across 60 design iterations, for $0.
IMPACT
$50K
in agency fees avoided
~2 weeks
vs. a three-month vendor timeline
Main nav
permanent slot in tennis.com navigation
2026 AI Impact Report · Joseph Yaksich
09 / 33
PROJECT 10
R&D / AI Infrastructure
Different Tennis Worlds
THE PROBLEM
Training our own models needed precise examples that would be slow and expensive to label by hand.
WHAT I BUILT
Virtual tennis worlds, from simple Three.js courts to detailed Blender scenes with varied lighting, cameras, surfaces, and players.
IMPACT
Auto-labeled
training data generated, not hand-labeled
Sim → real
models transfer to real broadcasts
Local
built and trained on a DGX Spark
2026 AI Impact Report · Joseph Yaksich
10 / 33
PROJECT 11
Programming / Digital Ops
Tennis Publishing Platform (TPP)
THE PROBLEM
There was no editorial layer between Gracenote and our apps: whatever came in from the provider went straight out.
WHAT I BUILT
An editorial checkpoint that ingests upstream updates, keeps our overrides, passes artwork from series to episodes, and publishes via XMLTV.
IMPACT
Editorial
control over what reaches our apps and partners
Persistent
overrides survive every upstream update
Automated
feed keeps running, with no repeat fixes at the source
2026 AI Impact Report · Joseph Yaksich
11 / 33
PROJECT 12
Digital Ops / Distribution
Tennis EPG Tool
THE PROBLEM
Gracenote View, used to QC channel metadata sent to Amazon, Roku, Samsung and others, cost $12K a year.
WHAT I BUILT
A self-hosted replacement built in under a week: a multi-channel weekly grid with artwork previews and role-based access.
IMPACT
$12K/yr
Gracenote View license eliminated
< 1 week
from start to working replacement
Self-hosted
on Tennis Channel infrastructure
2026 AI Impact Report · Joseph Yaksich
12 / 33
PROJECT 13
Content / Digital Ops / Analytics
BugNet
THE PROBLEM
Organizing a match by point means reading the on-screen scoreboard, and a general-purpose model was too slow.
WHAT I BUILT
A small, specialized model trained on synthetic and real scoreboards, built for speed on broadcast video.
IMPACT
3 → 154 fps
processing speed, roughly 50× faster
17 of 20
matches agree with reference point totals
Local
trained on a DGX Spark
2026 AI Impact Report · Joseph Yaksich
13 / 33
PROJECT 14
Consumer / Broadcast
The Learner’s Feed
THE PROBLEM
Traditional tennis commentary assumes viewers already understand the sport.
WHAT I BUILT
A prototype that combines AI video analysis, beginner-friendly scripts, and generated speech to explain each point in sync with the video.
IMPACT
Working
prototype, synced point by point
Interactive
viewers can ask questions as they watch
2026 AI Impact Report · Joseph Yaksich
14 / 33
PROJECT 15
Distribution / Apps
Amazon MCM Schema Tool
THE PROBLEM
Amazon VOD requires strict MovieLabs MMC/MEC XML packages, which operators were hand-editing asset by asset.
WHAT I BUILT
Co-developed a visual XML generator and API layer that standardized delivery and removed per-asset text editing.
IMPACT
Ended
schema-rejection loops with a zero-tolerance partner
~5 hrs/wk
of spreadsheet upkeep and daily email removed
Launch
expedited; still supporting delivery today
2026 AI Impact Report · Joseph Yaksich
15 / 33
PROJECT 16
Distribution / Amazon Ops
Amazon Channels Verification
THE PROBLEM
Checking live metadata against air schedules on 4 global feeds took daily spreadsheet updates and constant email.
WHAT I BUILT
Four real-time web dashboards (T2, TCDE, TCES, TCUK) that now serve as Amazon’s live source of truth for our schedules.
IMPACT
~250 hrs/yr
of spreadsheet upkeep and daily email ended
4 feeds
audited live, across global markets
2026 AI Impact Report · Joseph Yaksich
16 / 33
PROJECT 17
Editorial / Design
Gracenote Image Toolkit Generator
THE PROBLEM
Designers faced a two-week backlog producing multi-size artwork toolkits for ~65 ATP tournaments a year.
WHAT I BUILT
A generator that ingests the order of play and outputs customized, multi-format tournament artwork automatically.
IMPACT
~30 sec
per toolkit, down from a two-week backlog
195+ hrs/yr
of designer time returned
~65
ATP tournaments covered each year
2026 AI Impact Report · Joseph Yaksich
17 / 33
PROJECT 18
Executive
Jeff Snapshot Dashboard
THE PROBLEM
Jeff needed a reliable, recurring metric snapshot with reconciled channel and viewership data.
WHAT I BUILT
A dashboard that reconciles channel metrics with CourtSight data and automatically delivers verified snapshot reports.
IMPACT
Automatic
recurring delivery to leadership
Verified
channel metrics reconciled against CourtSight
2026 AI Impact Report · Joseph Yaksich
18 / 33
PROJECT 19
Digital Ops / Analytics
Game, Set, Metrics
THE PROBLEM
FAST viewership reporting was a manual, 14-step workflow with a three-week data lag.
WHAT I BUILT
An automated, always-on ETL pipeline: a Python email parser feeding SQLite/Snowflake and Metabase dashboards.
IMPACT
14 → 4
manual reporting steps
3 wk → 24 hr
data lag, from capture to dashboard
Since 2023
in continuous daily operation
2026 AI Impact Report · Joseph Yaksich
19 / 33
PROJECT 20
The Big T / Marketing
Podcast Analytics Platform
THE PROBLEM
Vendors pitched an $8K/yr tool that only estimated metrics by stitching separate web portals together.
WHAT I BUILT
I found the data was directly accessible and, in days, unified Amagi, Megaphone, Podstock, and Apple Podcasts in one automated dashboard.
IMPACT
$8K/yr
vendor tool avoided
4 → 1
platforms unified in one dashboard
In use
regularly by The Big T producers
2026 AI Impact Report · Joseph Yaksich
20 / 33
PROJECT 21
Editorial
Image Automation Suite
THE PROBLEM
After layoffs, a smaller team still had to produce daily head-to-head match thumbnails for the TC and PBTV apps.
WHAT I BUILT
Automated thumbnail compositing tools for both app properties, producing each match asset in seconds.
IMPACT
Seconds
per asset
Full
daily throughput kept with a smaller team
2 apps
Tennis Channel and PBTV
2026 AI Impact Report · Joseph Yaksich
21 / 33
PROJECT 22
Content / Video Analytics
CourtNet
THE PROBLEM
Finding the court in broadcast footage would have meant hand-labeling thousands of images.
WHAT I BUILT
A custom detector, trained on synthetic courts, that recognizes court geometry in real matches. Every later model starts from it.
IMPACT
Foundation
the starting point for all match analysis
Local
trained on a DGX Spark
2026 AI Impact Report · Joseph Yaksich
22 / 33
PROJECT 23
Content / Video Analytics
PlayerNet
THE PROBLEM
A visible court doesn’t mean play is happening, and generic detectors also catch officials and ball kids.
WHAT I BUILT
Adapted a player detector and combined it with court-position rules so only the players on court count.
IMPACT
293 / 293
marked singles and doubles points retained, with CourtNet
Local
trained on a DGX Spark
2026 AI Impact Report · Joseph Yaksich
23 / 33
PROJECT 24
Content / Editorial / Analytics
Replay Detection
THE PROBLEM
Replays look like real points, so the same action can be counted twice.
WHAT I BUILT
Combined player-movement matching with scoreboard visibility to flag repeated action automatically.
IMPACT
16 → 1
labeled replays left in the output
147 / 147
marked singles points preserved
2026 AI Impact Report · Joseph Yaksich
24 / 33
PROJECT 25
Platform
Local LLM Infrastructure
THE PROBLEM
We needed natural-language SQL before corporate AI tooling or zero-data-retention policies existed.
WHAT I BUILT
Deployed local open-weight LLM infrastructure (vLLM/Ollama API layer) for secure, offline natural-language queries over viewership data.
IMPACT
Offline
no viewership data exposed to the cloud
First
AI querying in place before corporate tooling
2026 AI Impact Report · Joseph Yaksich
25 / 33
PROJECT 26
Distribution
XML EPG Parser
THE PROBLEM
Triaging schedule issues across 85+ partner feeds meant reading raw XML line by line: 30–60 minutes per issue.
WHAT I BUILT
A visual feed parser that converts raw XML into a clean interface, turning code-level triage into a quick visual scan.
IMPACT
Seconds
per issue, down from 30–60 minutes
~75 hrs/yr
of triage time saved
85+
distribution partner feeds covered
2026 AI Impact Report · Joseph Yaksich
26 / 33
PROJECT 27
Distribution
Multi-Channel EPG Verification
THE PROBLEM
Overlaps, gaps, and metadata mismatches were caught only by ad-hoc spot checks, risking partner feed rejections.
WHAT I BUILT
A pre-distribution validation layer that audits every channel feed for timing conflicts and metadata errors before delivery.
IMPACT
Pre-delivery
errors caught before partners see them
All feeds
audited automatically, not spot-checked
2026 AI Impact Report · Joseph Yaksich
27 / 33
PROJECT 28
Content / Broadcast Ops / Analytics
AudioNet
THE PROBLEM
Video alone struggles to tell live play from players standing around between points.
WHAT I BUILT
A model that combines sound, movement, and player-position signals to identify rallies and separate play from downtime.
IMPACT
3 signals
audio, movement, and player position
Rallies
another independent way to find live play
2026 AI Impact Report · Joseph Yaksich
28 / 33
PROJECT 29
Content / Broadcast Ops
SpeakerNet
THE PROBLEM
Could announcers’ speech help separate live play from replays and breaks?
WHAT I BUILT
A speech timeline of the commentary, tested against labeled footage to see where it helps and where it misleads.
IMPACT
Mapped
where commentary is a useful signal
Protected
real points from an overly aggressive filter
2026 AI Impact Report · Joseph Yaksich
29 / 33
PROJECT 30
Finance / Content
Sinclair / Freevee Report Parser
THE PROBLEM
Finance and Content spent hours copying data out of Sinclair PDF remittances and Freevee Excel royalty statements.
WHAT I BUILT
An automated parser that turns PDF remittances and multi-format Excel files into one structured, queryable SQLite database.
IMPACT
Automated
ingestion replaces manual copying
Queryable
remittance and royalty data in one database
2026 AI Impact Report · Joseph Yaksich
30 / 33
PROJECT 31
Editorial / Content
S3-to-Aspera Video Delivery
THE PROBLEM
Hundreds of videos had to move from S3 to Aspera with no missed files, no duplicates, and limited local storage.
WHAT I BUILT
A restart-safe pipeline that skips existing files, batches downloads, verifies exact sizes, and deletes local copies only after checks.
IMPACT
540+
videos delivered
Repeatable
for every subsequent batch
2026 AI Impact Report · Joseph Yaksich
31 / 33
PROJECT 32
Programming Ops
Order-of-Play Parser
THE PROBLEM
Programmers hand-copied schedules from daily PDFs and converted local match times into the master planning sheet.
WHAT I BUILT
A watchdog service that parses each new order-of-play PDF, converts times to Pacific, and fills in the master sheet.
IMPACT
Automatic
from PDF to master sheet, no retyping
PT
local match times normalized to Pacific
2026 AI Impact Report · Joseph Yaksich
32 / 33
PROJECT 33
Executive / Programming
YoY China Open Report
THE PROBLEM
Tournament performance lived in static PDF reports with no way to drill down or compare years.
WHAT I BUILT
Turned the static PDF metrics into an interactive year-over-year viewership tool leadership can explore on its own.
IMPACT
PDF → live
static reports became an interactive tool
YoY
drill-down comparison of tournament performance
2026 AI Impact Report · Joseph Yaksich
33 / 33