The Radar
Wednesday, 1 April 2026
Today's picks
TRL v1.0
AI InfrastructureUnified post-training stack for SFT, reward modeling, DPO, and GRPO workflows.
Hugging Face just standardised the messy world of post-training with a production-ready framework. This is the PyTorch moment for model alignment, turning research experiments into reproducible engineering. Every team doing RLHF will want this.
Veo 3.1 Lite
AI PlatformsLow cost, high speed video generation via the Gemini API.
Google finally solved the video generation pricing problem with a model that cuts costs by more than half. This could be the breakthrough that moves AI video from demos to actual products. The race to the bottom on video pricing just accelerated.
Also on the radar
LFM2.5-350M
AI ResearchLiquid AI is proving that intelligence density matters more than parameter count. This tiny model challenges the scaling orthodoxy with better training rather than bigger weights. Smart move when everyone else is chasing gigantic models.
Pardus Browser
AI AgentsFinally, someone built a browser that agents can actually use without the overhead of human-focused features. This could become essential infrastructure as agents move beyond text into web automation. The anti-Chromium approach is refreshing.
CAI Cybersecurity Framework
AI SecurityA proper framework for cybersecurity agents with the guardrails and handoffs that actually matter in production. Most agent frameworks ignore security completely, so this fills a real gap. The multi-agent orchestration looks solid.
Hacker News
Show HN: Pardus Browser- a browser for AI agents without Chromium
16 pts 8 commentsA purpose-built browser for AI agents that strips away human-centric features. Could become essential infrastructure as agents move into web automation tasks.
Show HN: CargoWall – eBPF Firewall for GitHub Actions
14 pts 2 commentsAn eBPF-based firewall specifically designed for GitHub Actions. Addresses security concerns in CI/CD pipelines with kernel-level monitoring.
Reverse engineering GTA San Andreas with autonomous LLM agents
9 pts 1 commentsDemonstration of LLM agents working together to reverse engineer game code. Shows the potential for AI in complex software analysis tasks.
Show HN: DeepTable – an API that converts messy Excel files into structured data
7 pts 0 commentsAPI service that transforms unstructured Excel data into clean, structured formats. Tackles the universal problem of messy spreadsheet data ingestion.
AI Agent virtual cards with Privacy
6 pts 1 commentsVirtual payment cards designed for AI agents to make autonomous purchases. Privacy.com extends their service into the agent economy.