@a3rodev profile picture
Aero @a3rodev

dunno

Joined January 2025
  • Tweets 1,143
  • Following 68
  • Followers 29
74 Photos and videos
Pinned Tweet
Aero@a3rodev
May 24
Most meeting note-takers are either bloated or overpriced. Plus, the VPS hosting the @gobitsnbytes bot has 512MB RAM and 1 vCPU. So we built our own transcriber using @GeminiApp @Google @GoogleDeepMind Gemini 3.5 Flash. Here is how we got it running on a micro-server: (1/8)
9
2
6
145
Aero@a3rodev
Jun 10
not being able to use claude fable because I'm broke is giving me so much fomo
12
Aero retweeted
kartik@notKartikk
Jun 10
Claude
@claudeai
Jun 9
Introducing Claude Fable 5: a Mythos-class model that we’ve made safe for general use. Its capabilities exceed those of any model we’ve ever made generally available.
0:20
3
1
11
409
Aero@a3rodev
Jun 9
Genuine question for @OlaElectric: If an Ola S1 Air parked for ~2 weeks can go into “deep discharge” and need an ₹80k battery, with no warranty support, what does your 8‑year battery warranty actually protect? Our scooter (UP70GX4881) is dead at ~4,000 km.
2
79
Aero@a3rodev
Jun 4
SOTA intelligence 🥀
7
Aero@a3rodev
Jun 3
shipping glitchy ui because why not
4
Aero@a3rodev
Jun 3
ah.... even the YT comment section is so cooked.
3
Aero@a3rodev
Jun 1
claude.ai/public/artifacts/4… peak 2026

The Modern ABCs: 2026 AI & Tech Glossary

Explore 26 essential AI and tech terms defining 2026. A witty, interactive glossary covering agents, LLMs, embeddings, and modern development concepts.

claude.ai
5
Aero@a3rodev
Jun 1
kabir chillar my goat
193
Aero@a3rodev
Jun 1
9
Aero@a3rodev
May 31
1,373 commits this year. ~4 a day, every day.
1
3
74
Aero retweeted
Sam ☕
@samirande_
May 31
Am I the only person who doesn't watch cricket
61
1
105
5,655
Aero@a3rodev
May 31
we've no foundational model that can compete with the latest from us and china, whatever we've compares to their 2024-2025 model that too in indic language 🥀
9
Aero@a3rodev
May 29
bruh
12
Aero@a3rodev
May 28
an AI agent just emailed me a real $10 visa card. it runs on agentmail, the email inbox api for ai agents. cool stuff
1
3
485
Aero@a3rodev
May 27
fyi: @antigravity in the new 2.0 version zooming out works but zooming in doesn't work.....
29
Aero@a3rodev
May 24
Most meeting note-takers are either bloated or overpriced. Plus, the VPS hosting the @gobitsnbytes bot has 512MB RAM and 1 vCPU. So we built our own transcriber using @GeminiApp @Google @GoogleDeepMind Gemini 3.5 Flash. Here is how we got it running on a micro-server: (1/8)
9
2
6
145
Aero@a3rodev
May 24
The code is live, pushed to main, and passing all 181 unit tests. Turns out you don't need expensive SaaS integrations if you leverage basic streams, local FFmpeg, and cloud-side models. (8/8)
2
34
Aero@a3rodev
May 24
Replying to @gobitsnbytes @GeminiApp @Google @GoogleDeepMind
We don't keep any audio. Everything is deleted post-processing. Transcripts are stored in SQLite. Attendees can retrieve them anytime using /meet-transcript, but the access control is tight—you can only pull meetings you actually attended. (7/8)
2
36
Aero@a3rodev
May 24
Replying to @gobitsnbytes @GeminiApp @Google @GoogleDeepMind
The merged audio goes to Google's File API, and Gemini 3.5 Flash generates a structured JSON output (summary, key decisions, action items). The bot then DMs every attendee the notes and attaches the full transcript as a .txt file. (6/8)
2
43
Aero@a3rodev
May 24
Replying to @gobitsnbytes @GeminiApp @Google @GoogleDeepMind
Once the voice channel is empty, the bot leaves. A local FFmpeg process merges the per-user files using their join/leave timestamps as offsets to keep the conversation in sync. We run this in a sequential queue so the 1 vCPU doesn't choke. (5/8)
2
34
Aero@a3rodev
May 24
Replying to @gobitsnbytes @GeminiApp @Google @GoogleDeepMind
Since we speak a mix of English, Hindi, and Hinglish, we prompt Gemini to transcribe in whatever language is spoken, but summarize in English. We also added a quick /hindi text command in the VC chat to trigger the Hindi voice notice on demand. (4/8)
2
59
Load more