I build AI agents that call tools instead of guessing, plus the Rust and full-stack systems around them. 5th of 362 at ICML 2026 repro. Electronic + math nerd.
Between the PCB and the prompt
Joined January 2012
- Tweets754
- Following206
- Followers24
- Likes120
Codex CLI feature request: let long shell commands run in the background and wake the agent when they finish. Today a 2h build means calling wait every 60s, each call re-sending ~150k tokens of context. One hour: 337 model calls, 98 just waiting, 55M input tokens. @thsottiaux
Dinesh Jinjala retweeted
Some Jev Demos worth to have a look
Virality / Content
Viral post scorer (SuperX) — @robj3d3 — nitter.cf/robj3d3/status/2100722…
X content filter extension — @marcelpociot — nitter.cf/marcelpociot/status/21…
Feed optimizer — @redp314 — nitter.cf/redp314/status/2100915…
Personalized Hacker News ranking — @VisheshBaghell — nitter.cf/VisheshBaghell/status/…
Trading
On-chain trading bot (Monad/Kuru, 300ms) — @jarrodwatts — nitter.cf/jarrodwatts/status/210…
Trading engine + price-path model — @BrendanPlayford — nitter.cf/BrendanPlayford/status…
Price-action signal detector — @HumeTrader — nitter.cf/HumeTrader/status/2100…
Multi-signal trading calls — cited by @KevinMagnan (@_trou3) — nitter.cf/KevinMagnan/status/210…
Browser / Agents
Browser use + Stagehand — @kylejeong — nitter.cf/kylejeong/status/21006…
Google Flights ~7.1s (Jev Ultrafast) — @tonysimons — nitter.cf/tonysimons_/status/210…
Chrome agentic browsing extension — @razaanstha — nitter.cf/razaanstha/status/2100…
Live Excalidraw drawing agent — @VladTerin — nitter.cf/VladTerin/status/21007…
OpenCode browser CLI / UI tests — cited by @KevinMagnan — nitter.cf/KevinMagnan/status/210…
Agent memory compaction — @kylejeong — nitter.cf/kylejeong/status/21008…
Coding / Agent Control
Codebase overengineering classifier — @ryanvogel — nitter.cf/ryanvogel/status/21000…
Local coding agent done-gate — @matrixoar — nitter.cf/matrixoar/status/21005…
Foreman (watch coding agents) — @JoshARosen — nitter.cf/JoshARosen/status/2100…
Risky tool-call gate — @alg0agent — nitter.cf/alg0agent/status/21009…
Lead-decision latency benchmark — @MejiasDev — nitter.cf/MejiasDev/status/21009…
Vercel fx command review — cited by @KevinMagnan — nitter.cf/KevinMagnan/status/210…
Search engine + Jev Review Action — @fatwang2ai — nitter.cf/fatwang2ai/status/2100…
Chat / Moderation
Realtime chat moderation — @stoufax — nitter.cf/stoufax/status/2100908…
Tool-calling chat bot (no LLM text) — @CodingGarden — nitter.cf/CodingGarden/status/21…
Matching / Ranking
Reranker vs Voyage/Cohere — @aHev — nitter.cf/aHev/status/2100692958…
Job-apply extension — @vammu920 — nitter.cf/vammu920/status/210059…
Resume↔job + auto-parts matching — cited by @KevinMagnan — nitter.cf/KevinMagnan/status/210…
Games / Sims
Wordshift semantic typing racer — @marcelpociot — nitter.cf/marcelpociot/status/21…
Tetris early demo — @marcus_lowe — nitter.cf/marcus_lowe/status/210…
Tetris typed move judgments — @DineshDataAI — nitter.cf/DineshDataAI/status/21…
Tetris live Choice demo — @mirzaaghazadeh — nitter.cf/mirzaaghazadeh/status/…
40-person town sim — @threepointone — nitter.cf/threepointone/status/2…
JFK airport manager sim — @alessandro_a0 — nitter.cf/alessandro_a0/status/2…
Dota 2 smurf detector — @MenilVukovic — nitter.cf/MenilVukovic/status/21…
3D forest NPC behavior — @Cephalization — nitter.cf/Cephalization/status/2…
Chakravyuha pathfinding — @reachmeviz — nitter.cf/reachmeviz/status/2100…
Creative / Other
Visual reference finder — @albicodes — nitter.cf/albicodes/status/21007…
Text → emoji sentiment — @ColinMcDermott — nitter.cf/ColinMcDermott/status/…
Smart-home room settings — @moelabs_dev — nitter.cf/moelabs_dev/status/210…
Website → native app (Shipper) — @chddaniel — nitter.cf/chddaniel/status/21009…
Real estate photo cost cut (Realisto) — @apagut — nitter.cf/apagut/status/21009175…
Meta
Roundup thread — @KevinMagnan — nitter.cf/KevinMagnan/status/210…
Jev project tracker — @timmyships — nitter.cf/timmyships/status/2100…
in 2026, why do you still get irrelevant posts on your X feed?
built a feed optimizer with @typesafeai Jev.
it's tuned on my own bookmarks, then every post goes through ONE call to Jev → 9 typed answers as probabilities: ragebait, ad, low effort, hostile, useful… and "would i save this?"
code turns that into HIDE / SKIM / KEEP. 150 real posts from my For You in the video, $0.015 for all of them, the whole feed scored in under 10 seconds
tested on 300 bookmarks it never saw: 63% land in KEEP. no LLM, nothing trained, 16 of my saves go in as examples and the policy is a few lines of code
I built a Tetris demo with Jev.
It returns a typed judgment:
→ rotate once
→ place at column 7
→ drop
The game engine checks the move and locks it in.
Meanwhile, I still rotate the square piece three times “just to see.” 😄
@typesafeai
I was using Luna on Ultra but after some time in new session again I only see till Max. @thsottiaux
Your AI agent has DB access. Someone asks it to export customer emails.
A prompt saying "don't" is a sticky note on the server-room door.
Built for @WeMakeDevs × @truefoundry, reviewed by @Qodo: an agent that can't leak, the export tool doesn't exist. 🧵
Favourite catch of the week: @Qodo's review flagged that a sentence saying the digest does NOT equal the hash would have PASSED our verification gate we only checked the hash appeared. Real finding, one-line fix, public thread in the repo.
Repo clone it and verify a release receipt yourself in 60 seconds:
github.com/MachineLearning-N…
Built on TrueForge.
Dinesh Jinjala retweeted
Linus Torvalds says tracking down a bug in Linux's Intel GFX Driver was a "debug session from hell." The AI helping him repeatedly called it "impossible and unsolvable" and suggested they just write a report instead.
He kept pushing. 24 debug patches and 18 kernel boots later he found the bug: one round_up() that should have been round_down().
He then let the AI write the commit message.
git.kernel.org/pub/scm/linux…
Dinesh Jinjala retweeted
arXiv -> quickarXiv
Papers are often written in convoluted language that is hard to understand
Swap arXiv -> quickarXiv on any paper URL to get an instant blog with figures, insights, and explanations. Now extracted with @Zai_org's GLM OCR 🚀
Also includes code + author tweet
Dinesh Jinjala retweeted
Omarchy Mac Quattro is here!
- Quickshell + new theming
- Lua-based Hyprland config
- Full disk encryption
- Improved Apple Silicon support
- Btrfs + Snapper snapshots
- Faster, one step installation
🔗 github.com/omarchy-mac/omarc…
@dhh @tiredkebab @bi9kahuna
I just love GLM-5.3 @Zai_org great work and great model. More clear and understandable words.
Dinesh Jinjala retweeted
Thank you! Pi is many DeepSeek researchers and developers favorite daily drive.
DSH reused Pi's LLM adaptor package for connecting to non-DeepSeek models and it was a great experience that just work.
Excited to join the global Open Source agent harness community!
Dinesh Jinjala retweeted
Introducing GLM-5.3: Built to Code. Ready for Cyber Defense.
- Top-tier coding and agentic capabilities, achieved through post-training on the 743B base model
- A major leap in cybersecurity, setting a new standard among open models
Tech Blog: z.ai/blog/glm-5.3
I built an open-source project called SQLite Graph Accelerator.
It helps SQLite work with graph data like people and connections.
It supports Python, graph searches, paths, composite keys, and safe results when data changes.
In one test, two-level graph searches were about 3.97× faster.
github.com/MachineLearning-N…
#SQLite #Rust #OpenSource
GPT-5.6-Luna is at capacity. Not able to use that model right now(On Pro plan). Anything in future to increase model capacity for luna? @thsottiaux @sama
Dinesh Jinjala retweeted
Finished at top#3 for @icmlconf reproduce challenge hosted by @askalphaxiv on @huggingface !
It was super fun and furthered my knowledge.
My old friend @gamiqaq was really helpful with powerful insight.
