I just saw one of the most practical in-car AI assistants. AutoNeural-VL-1.5B is the first production-ready, real-time multimodal AI model built for the Qualcomm SA8295P NPU, and it seriously raises the bar for in-car AI.

Who all am I going to meet at the @APIdaysGlobal Paris conference? 👋

Almost after two months of using iPhone 17 Pro Max: > battery is phenomenal > no heating issues at all > camera is great in daytime. Night mode may need some improvements > speaker is insanely loud > screen is so smooth I think the improvements are so visible to me because I

2025 AI recap (not in order): DeepSeek V3.2, Opus 4.5, Gemini 3, GPT-5, Grok 4, Qwen3-Next, Qwen3-Omni, Claude 4, Llama 4, Runway Gen-4, Gen-4.5, Mistral Large 3, Codestral, DeepSeek R1, Helix, Manus What did I miss?

Has the launch of DeepSeek V3.2 failed?

Most smart people I know don’t try to know everything. They go deep in their domain, and when something outside it comes up, they simply say “I don’t know” and move on. The real skill is not trying to learn everything.

This could be your next engineer hire in 2026: • v0 • Bolt • Claude • Cursor • Replit AI • ChatGPT • Windsurf • Kiro • Warp • Blackbox • Copilot • Tabnine • Lovable • Emergent • Antigravity • Anything Max • Continue • Supermaven • GitHub Spark What else?

Am I the only one who uses Apple Music instead of Spotify? 👀

Perfectionism without shipping just means you clearly see why nothing gets done. done > perfect

pre-2023, no one expected us to ship in 48 hours. The entire pace and expectation of software was different. For years, we all knew that coding was a slow process. If we didn’t know something, we sat with it. We read docs, we looked for resources, we asked colleagues, we

Whenever I see programmer salary numbers from India, it suddenly makes sense why brain drain is such a big thing here.

I just saw one of the most practical in-car AI assistants. AutoNeural-VL-1.5B is the first production-ready, real-time multimodal AI model built for the Qualcomm SA8295P NPU, and it seriously raises the bar for in-car AI.

@nexa_ai If you’re curious about the future of smart cars, keep an eye on this. Check out the official Git repo: https://t.co/lmfJBGWw0T Appreciate the team for collaborating with me on this post.

@nexa_ai 3. Decoding different signs

@nexa_ai 2. In-cabin safety

@nexa_ai Check out these couple of use cases. 1. Finding a parking slot.

@nexa_ai And everything runs on the NPU, not the cloud. So even with zero network: • it still reads signs • it still detects danger • it still helps you navigate • it keeps your privacy inside the car

@nexa_ai The coolest part: vision + language together. Example: You get a message on CarPlay → AI reads it → understands the plan → starts navigation → replies with your ETA. No cloud. No wait.

@nexa_ai It can see the inside of the car clearly. High-resolution vision lets it: • notice a child moving • spot if something falls • check seatbelts • read signs • understand driver behavior

Most “car AI” is slow or depends on the internet. AutoNeural-VL-1.5B by @nexa_ai is different. It’s already running inside real cars, works offline, and reacts almost instantly. The latency is under 100ms. This means the AI understands things as they happen.