adi (@asahas) header
adi (@asahas) profile picture

engineering @pointer, partner @dormroomfund

Joined May 2023
Pinned Tweet
Can confirm theres open desks 👀
Pointer is hiring! We're growing faster than we can handle and need people across every function. We build AI systems that do a company's hardest work and prove it was done right, deployed in some of the highest-stakes operations in the world. If this sounds interesting, learn more at pointer.ai/careers. Refer someone we hire and we'll pay you $15,000.
2
82
162,321
No APIs. Tons of different documents (Lens!). An accuracy bar at 100%. And it has to run identically on day one to day one thousand. Great working with Torfs on this!
Torfs is Belgium’s largest shoe retailer: 3 million pairs a year, across 15,000 different products. For decades their team moved it all between systems by hand. RPA broke on it. Now they wake up to finished work, every day, on Pointer.
3
7
423
Excited to share Lens, our new document engine. It’s the most accurate dedicated extraction system on leading document benchmarks, and it sits on the Pareto frontier of cost and accuracy. 🧵 1/4
5
5
3
13
1,363
Lens treats every model output as a draft. Code maps the page, a model reads it, and code checks the answer against the page. For example, it counts the rows and adds up the totals, and anything that doesn’t tie out gets read again. On ExtractBench Pro, Lens got 63 of 100 hard documents 100% right, the top score on the leaderboard. For reference, Codex GPT-6 Astra got 58, at roughly 8x the cost per page. 🧵 3/4
1
3
108
Today we're launching Boston Defense! We're an independent student organization where engineers, researchers and philosophers meet to advance defense technology. Our first two projects are already underway, a mechanically scanned radar system that detects and tracks small drones, and our own autonomous UAS swarm. In January 2027, our team is heading up to New Hampshire together to take both into the snow and see how they hold up in freezing conditions. Currently recruiting and looking for partners and sponsors. DM me if you're interested! Links below!
26
13
1
136
7,414
The only place where you’ll see Jasper O’Farrell mentioned in an engineering blog Come join us
Pointer is hiring! We're growing faster than we can handle and need people across every function. We build AI systems that do a company's hardest work and prove it was done right, deployed in some of the highest-stakes operations in the world. If this sounds interesting, learn more at pointer.ai/careers. Refer someone we hire and we'll pay you $15,000.
2
1
30
7,178
We at @pointer hosted a dinner with @stepstonegroup at Le Pavillon in NYC last week for a group of finance leaders in banking, insurance, fund admin, and more. It's clear that nobody wants AI they have to babysit. The systems delivering the most value are the ones that can remove the human reviewer entirely and run in the background for hours. More soon!
2
7
19
1,272
adi retweeted
locked in the trenches coding the 2 most requested features for the roommate app: undo button for when your fat fingers accidentally skip a match (the animation is so smooth) “see more” button to read their full bio link in bio 😉
1
2
217
adi retweeted
random roommate matching is literally russian roulette lmfao. we looked at the data on what actually destroys shared housing. the results are honestly kind of wild. if u want to avoid a 12-month lease from hell, look at these numbers:
1
1
2
174
Great work with many nuggets of practical wisdom. And one of the best technical blog posts I've seen in a while!
Today, we’re sharing a new state of the art for computer use. Our system holds the two highest verified scores on OSWorld, the standard benchmark for AI agents that operate a computer like a person: 83.6% using Claude Opus 4.7 and 81.5% using Claude Sonnet 4.6. The human baseline is 72.4%. 🧵 1/7
1
2
5
1,982
A lot of people have been asking about our harness / approach - some thoughts: 1/ it’s fully open source on github! 2/ it is quite simple - and we think this is where harness engineering is heading. you no longer need elaborate scaffolding to force the model to reason in a prescribed way 3/ we initially included a verifier to check the executor’s work. it ended up being *more* accurate than the benchmark’s grader, but omitted it (you can't score above the ceiling set by the grader). we have a lot more to say on this. 4/ we were most excited by the performance uplift in sonnet (lighter model). it reflects a shift toward picking the model at the intelligence/cost pareto max for a task, not just the largest one. sonnet achieved near parity with opus in performance, while costing less than half.
Today, we’re sharing a new state of the art for computer use. Our system holds the two highest verified scores on OSWorld, the standard benchmark for AI agents that operate a computer like a person: 83.6% using Claude Opus 4.7 and 81.5% using Claude Sonnet 4.6. The human baseline is 72.4%. 🧵 1/7
10
15
196
43,581
Today, we’re sharing a new state of the art for computer use. Our system holds the two highest verified scores on OSWorld, the standard benchmark for AI agents that operate a computer like a person: 83.6% using Claude Opus 4.7 and 81.5% using Claude Sonnet 4.6. The human baseline is 72.4%. 🧵 1/7
52
41
10
284
187,162
if you’re a cracked student founder in boston go fill out bos.network we’re making a directory for people like you
1
1
1
6
1,160
finally fixed alot of the bugs with the editor and implemented liveshare for videos. will be live soon ‼️‼️
2
2
14
1,383
Better view of live editor timeline in action
5
545
decided it’ll be easier to launch the video editor and live collaboration together, so i’m just pushing for that this week. made a ton of progress today but still got some bugs to fix. attached screenshots on my progress! very exciting
2
13
1,005
ai plants
1
1
7
1,570