Two product lines
Built in Islamabad, run in production.
Why it matters
A DRS day costs $21,000 before anyone travels.
Traditional decision-review and broadcast analysis brings 15 to 20 people to the ground: camera operators, analysts and a truck, moving with the fixture. Day rates run $20,000 to $30,000 before logistics. That is why the technology only reaches the top tier of the game.
Chrio replaces the crew with software. Detection, prediction, render and output run on a single workstation at the ground, operated by one person, for roughly 0.15x the cost. Domestic leagues, regional tournaments and academy cricket become viable broadcast clients.
- Traditional crew
- 15–20
- people on site
- With Chrio
- 1
- operator, one machine
- Day rate today
- $21k
- before logistics
- Logistics with Chrio
- $0
- no truck, no travel
The cricket products
One feed in. Three products out.
Every product shares the same detection and event pipeline, trained on a 30 TB dataset from PSL, National T20 and Tri-Nation cricket. Building one improves the others.
CritShot
LiveReads the delivery, the field and the game state, then renders the shot the batter should have played as a 3D avatar on the live broadcast, with a difficulty score for the call.
CritRate
In developmentA 1 to 10 rating for every player, every match, computed from in-game performance as it happens. No manual scoring, no panel of pundits.
CritDRS
In developmentDecision review for domestic cricket from the same tracked trajectory: pitching, impact and wickets adjudicated without a dedicated DRS crew. Ball tracking is built; collision detection and multi-view fusion are the current focus.
Dhundo · Live at dhundo.store
One search reads 136 Pakistani stores.
Most price-comparison services in Pakistan cover a fraction of that. The long tail is the point: the single-brand shops that only sell their own catalogue are where the price differences actually live, and they are the ones nobody else indexes.
Stores are added by an engineer who writes and registers a model for each one, then seeds its first results off-peak. There is no self-serve onboarding, which is what keeps the index clean.
What you actually get
Five stores, five prices, one rail.
Every result carries the store that quoted it and links straight out to that store's page. Dhundo holds no stock, sells nothing, and takes no payment — the transaction is between the buyer and the shop.
Results that nobody has rated say “No ratings yet” rather than showing a number somebody invented, and paid placements are labelled Sponsored where they sit.
The part nobody else publishes
“It is not a customer rating. Nobody has voted on it, and we do not publish it as though they had.”
Dhundo publishes a page whose entire purpose is to limit its own claims — including a section titled “What cannot move the score”.
Fake ratings are banned, and a test enforces it
Review and AggregateRating structured data cannot ship. Most competitors emit synthesised aggregate ratings for search ranking. Dhundo refuses, and proves it in CI rather than asking to be trusted.
Price age is disclosed, not hidden
A price is what the store showed when Dhundo last read it, not a live lookup. The product says so plainly and tells you to confirm before paying.
Paying does not buy a ranking position
Sponsored placements are labelled where they appear, and money does not move a result up the list.
How it holds the latency
Total time is the slowest store, not the sum of all of them.
One query fans out to 136 store adapters across three independent scraper fleets, and results stream back as they arrive. What comes back is then reranked in two stages in production: a bi-encoder for recall, then a cross-encoder for precision.
That is the same shape as the cricket work — fan out across many sources in parallel, reconcile what comes back, answer under a latency constraint. CritShot runs YOLOv8, YOLOv11 and Detectron2 as a parallel consensus ring with SAHI recovery at 1.5 ms; Dhundo runs 136 store adapters across three fleets and then reranks. Two problems, one engineering pattern.
What it has done so far.
Read from Dhundo’s live stats. As of 3 September 2026.
- Rs 1.11B
- in price differences surfaced
- 51,612
- searches run
- 136
- stores read in parallel
- 966
- shoppers using it
How we build
Two products, one habit: run everything in parallel, then agree.
Cricket · detection
Three detectors see every frame. A ball is only a ball when the ring agrees; SAHI sliced inference recovers the ones too small for a full pass.
Dhundo · search
One query fans out to every store at once. Results are normalised and merged into a single price-sorted list before the first store would have finished on its own.
Where a library got in the way, we rewrote it: the only Python NDI library, with a critical memory leak removed, and an assembly-optimised WebSockets library.
Technology
Six stages, one workstation, broadcast out over NDI.
The camera feed comes in. Three detectors, YOLOv8, YOLOv11 and Detectron2, vote on every frame, and SAHI sliced inference recovers the balls too small for a full pass. An LSTM and XGBoost stack predicts the shot. Blender renders the avatar. The finished overlay leaves the machine as an NDI stream the broadcast truck already knows how to take.
Models are trained with TensorFlow and Roboflow and served on Nvidia hardware. Detection accuracy on cricket gameplay artifacts is 98.67%, up from 20.95% in the first model version in September 2024.
- 01Feed inBroadcast camera, SDI capture
- 02CNN consensusThree detectors agree
- 03SAHI recoverySliced inference for small balls
- 04LSTM · XGBoostShot prediction
- 05Blender render3D avatar in real time
- 06NDI outStraight to the broadcast
- Tooling credits, not partnerships.
Distribution
Trans Group and TPT carry CritShot to a national broadcast.
Trans Group is Pakistan's largest sports business group; Trans Production and Technologies is its production arm. The partnership takes CritShot from a review desk at a national-level test to the live feed millions of fans watch.
Chrio has been incubated by the National Incubation Center Islamabad, NSTP PITB and Invest2Innovate, whose support has carried R&D, funding readiness and product development.
In the field
Tested at the ground, not in a demo room.
CritShot ran at a national-level test in March 2025 with one operator running four feeds, game state and live metrics from the broadcast box. The team has since worked pitch-side, in exhibition booths and on competition stages.
Founders
Abdullah Zubair Ghouri
Co-founder, CEO. Started Chrio in May 2024 as a second-year computer science student at FAST-NUCES, raised the SAFE and signed the Trans Group partnership. Also worked on RL agents at Motive and LLM training at MAVN AI.
“Every game deserves technology that makes cricket more intelligent, transparent, and exciting. That is the future we are building at Chrio.”
Saffi Muhammad Hashir
Co-founder, CTO. Owns the technology end to end — the detection models, the real-time vision pipeline that holds 30 FPS on a single GPU, and the NDI layer that puts it on air. Cut inference from 200 ms to 33 ms; his research ranked first on an international benchmark at a twentieth of the compute. Leads 25 engineers.
“The best technology becomes part of the game itself. Our goal is to make powerful AI feel instant, reliable, and effortless.”
Recognition
- Google Startup ProgramWinner, $25K · Sept 2025
- National Ignite AI Wrapper AwardRunner-up, EdTech · Oct 2025
- HBL P@SHA ICT AwardsRunner-up, Media & Entertainment · Sept 2025
- Prime Minister Pitch PerfectFinalist, top 50 · Nov 2025
- Citadel National PathfinderTop 8 · Nov 2025
- Vision AI HackathonWinner, action detection · Jan 2025
Incubated at NIC Islamabad, NSTP PITB and Invest2Innovate.
Work with Chrio.
Abdullah Zubair Ghouri, CEO
abdullahzubair@chrio.siteSaffi Muhammad Hashir, CTO
saffimuhammadhashir@chrio.siteIslamabad
Chrio PVT LTD · Plot No. 24-B, Street No. 6, Sector H-9/1, Islamabad, Pakistan