Ashwith Basani
24 posts
Ashwith Basani
@BasaniAshwath
AI engineer and Solutions Architect. in pursuit of meaning to life. Building an esports analytics and coaching platform to help fps gamers and coaches.
Hyderabad Katılım Ağustos 2024
125 Takip Edilen24 Takipçiler
keep pulling up the ladder, I will pull it down for those who deserve it but can't access it :)
seth@sethsetse
you are ruining the system
English
@PlatoonVAL @patmenVLR Hey @PlatoonVAL, we at Clutchlabs AI are building Video analytics on pure gameplay video for Valorant teams, we are in beta stage right now working with few teams. Sent you detailed context in your dms, could you check and let me know if it is something that could help your team?
English
This is updated for 2026 (missing a couple recent games) and I swear no bias to @patmenVLR since he's my teammate 😅 - FYI the numbers from when I did it originally in 2024 were jacked up a bit because they counted assisted kills, but Leo was clearly still dominant then
leiysoar | cleiYo ♡@leiysoar
Who is the scariest person to peek in VCT Has to be Alfajer right??
English
ClutchLabs AI (clutchlabs.ai) - we're building video intelligence infrastructure for gaming and sports. Hours of match footage across both industries sit unused or deleted daily, while that data could power betting platforms, drive player development decisions, and automate performance tracking at scale. CV/ML founder, pipeline live on real footage, Currently holding an LOI with India's ENC national Valorant head coach.
English
When we got into YC Spring ‘26, we had 3 YC alum generously refer us.
It made all the difference.
I’d like to pay it forward :)
Pitch me what you’re building and why you’re the right team to solve it in 1-2 lines.
I will refer 10 founders.
English
@jai_mansukhanii Video Intelligence/Analytics for Gaming and Sports
English
got a good idea with traction?
I’ll refer you + 10k scout cheque
a16z speedrun 🧊@speedrun
a16z speedrun will return to San Francisco for our summer/fall cohort. Applications open tomorrow. Who's planning to apply?
English
@boardyai Video analytics/intelligence for Esports gaming. Currently holding an MVP and looking out for prospect conversations and customer signups, and investors
English
Ashwith Basani retweetledi
RT this + Signup on aigrants.in
= Lifetime perks + Unlimited tokens this @OfficialINDIAai week thx to donors
@AIGrantsIndia@aigrantsindia
Join the movement for special lifetime perks
English
@speedrun @andrewchen Is this global or just for people in USA?
English
NEW PROGRAM: If you're a student or recent grad, apply to the a16z speedrun Alpha Fellowship for
> A $20k grant (equity-free)
> Eligibility for up to $250K investment
> OR placement at one of our portcos
This is THE program for pre-idea, pre-team founders and technical talent.
andrew chen@andrewchen
a16z Speedrun Alpha, for pre-idea/pre-team/pre-everything founders it's time to bet on yourself, and figure out your startup idea. 2026 is well underway, crazy stuff happening in AI, and you're building agents/apps/whatever every night+weekend. You want to start a startup but you're working or still going to school. what if you're pre-idea, pre-product, pre-launch, and even a solo founder? You need time to cook The Alpha Fellowship is for you. alpha.a16zspeedrun.com details: - $20K equity-free upfront to start building - up to $250K investment when you finalize - automatic final interview for a16z speedrun, with up to $1M investment - 8-week, in-person experience with a kickoff retreat, founder AMAs, and small-group dinners alongside the a16z speedrun community - targeted to early-career highly technical founders - deadline to apply is March 6 We ALSO have a "startup track" for the Alpha Fellowship where you can get more founder experience by working for a portfolio company if you're not quite ready to found something. The Alpha Fellowship places top early-career engineers into full-time roles at fast-growing a16z speedrun and Andreessen Horowitz portfolio companies. For future founders, we provide capital before a team or idea even exists. We're looking for highly technical students and recent grads who don't want to wait to start building. Fellows take full-time roles at fast-growing portfolio companies - or, if you're ready to build now, receive capital to start your own company - kicking off with a two-month in-person fellowship. Fellows also have access to the a16z speedrun and EO Ventures communities and events. ... If this is you, want to meet you. If you have people to introduce us to, that would be amazing too. will have more to say, and lots of ideas coming up here. But excited to get this out! Excited to host y'all soon.
English
Day 27
Today and the following days will look gloomy with no appealing results, as I delve into the grunt work of getting the data annotated and iterating on the best model.
Same things, but getting better day by day.
#buildinpublic
English
Day 25
Today was all about R&D. Thanks to the @AWSstartups 5k $ cloud credit grant, the continuous R&D cycles have become stress free for me.
~1500 epochs across 10 models to find the right segmentation approach.
SAM: centered around prompt based usage, even with proper segmentation mask data, outputs single mask per click - wrong paradigm.
YOLO: instance segmentation, designed for multiple distinct objects - wrong paradigm.
Ensemble (U-Net + YOLO): overcomplicated, YOLO background_area hit 23.6% mAP.
U-Net won. Compared to latest models UNet's architecture is designed for the task, given smallest possible backbone resnet34 works pretty well - semantic segmentation is exactly what "classify each pixel" needs.
Trained on 13 samples -> generated predictions on 1,100 crops. Good enough to correct rather than annotate from scratch. V2 training next.
Excited and Nervous, as I approach the closing steps of my product backbone.
What's something geeky you're working on?
#buildinpublic #computervision
English
Day 24
Game videos feature varying map orientations. To extract consistent position patterns, minimaps must be reconstructed into a base coordinate system.
Initially tried contour matching, achieved 90.5% "high confidence," but visual checks revealed errors in most outputs.
Explored SuperPoint/SuperGlue (fails beyond 45° rotation), template matching (no rotation support), and HSV segmentation (too fragile).
Key insight: background and player icons act as noise. Extract clean map edges first, then match. Pivoting to segmentation-first approach.
Built a SAM assisted labeling backend for active learning; patches are ready, labeling starts tomorrow.
#buildinpublic #computervision
English
Day 23
Back after a week of break due to some unforeseen incidents. Scoped the MVP into multiple phases, worked out a campaign to reach out founding users. Built ML feasibility analysis pipeline.
#StartupIndia #StartupLife
English
Day 15
Non-tech update.
Moved cities. Joined an accelerator cohort. Pitched my startup.
Calling myself a founder used to feel fake.
Today, surrounded by other builders, it felt real for the first time.
English
Day 12
Been quiet for 2 weeks — traveled to Bangalore and back, life got in the way.
But still shipping:
→ Tested 3 models (YOLOv11n, RT-DETR L, RF-DETR) — went with RF-DETR (open source, commercial friendly, great on small objects)
→ Built the minimap position detector
Now packing up to move to a new city for 6 weeks for an accelerator bootcamp. Excited and nervous.
#gamingstartup #gaminganalytics #startupjourney
English
Ashwith Basani retweetledi
@victor_bigfield I got Aws credits from Aws activate program, and I’m experienced with cloud - so I’d pick Amazon rds for Postgres, and valkey for caching
English
Day 9: Turned two days of debugging into a blog post: "How I Annotated 60k+ Images Without Losing My Mind."
This is for everyone working on ML/DL models and data labeling across all domains.
Active learning with Label Studio (@LabelStudioHQ) + remote GPU.
Stumbling blocks: Python multiprocessing killing instance attributes, JWT refresh tokens that aren't access tokens, bidirectional tunnels via ngrok + SSH.
The patterns apply beyond computer vision—Label Studio supports 20+ task types across NLP, LLMs, speech, and OCR.
#buildinpublic #machinelearning #labelstudio #computervision
@ashwath.basani_83895/how-i-annotated-60k-images-without-losing-my-mind-building-an-active-learning-pipeline-with-label-245eeb3f41f7" target="_blank" rel="nofollow noopener">medium.com/@ashwath.basan…
English
Ashwith Basani retweetledi
Google Using Big Data to Make Blockchains Searchable
wittysparks.com/google-endeavo…
#bigdata #blockchain #technology
English
Day 8
Finally connected my local Label Studio to an EC2 GPU for model training via github.com/HumanSignal/la…
The setup: ngrok exposes the local app, and an SSH tunnel connects back to EC2's ML backend. Sounds simple until Python's multiprocessing 'spawn' ignores your carefully set instance attributes.
starter annotations ready. Just need to debug one more subprocess issue, then continuous training iterations begin right from the label studio.
I will be sharing a detailed article on how Label Studio wins on the current labelling tools for different model training tasks once I set it up successfully.
@humansignal_
#buildinpublic #machinelearning #labelstudio
English
Day 6
Extracted 60,847 minimap patches from gameplay videos and organized them into 308 annotation batches. Ground truth labels from 106 games across 7 maps.
Set up self-hosted Label Studio for annotation after frustrations with Roboflow's API permissions blocking uploads, buggy persistence, and costly subscriptions.
Next: Setting up Label Studio ML backend for active learning—annotate a few, let the model suggest the rest.
Started writing about things I've learned on this journey. Here's my first article on how to 10x your Claude code usage: @ashwath.basani_83895/claude-code-not-an-other-chatbot-an-ai-development-team-sdk-f3ca59002f94" target="_blank" rel="nofollow noopener">medium.com/@ashwath.basan…
#buildinpublic #computervision #gamedev
English
Day 4
Ran HUD detection on ~2k Valorant clips to generate training data for the next model (minimap player position tracking) on the cloud via S3.
Then shipped dev tooling: set up bidirectional sync between my journal repo and project workspace with a post-merge git hook that auto-syncs Claude Code skills and agents on every pull across workspaces and devices.
Infrastructure usually feels invisible, but it's the difference between shipping fast and drowning in coordination overhead.
#buildinpublic #machinelearning #devops
English