Quanta AI Labs

01 / WRITING

Notes from building it.

Long-form engineering writing on applied AI — what we build, what broke, and what we would do differently. No listicles.

  1. 01

    Jev cannot hallucinate. Its own docs explain the catch.

    TypeSafe AI raised $40M for a model that guarantees zero hallucinations. The guarantee covers schemas, not answers, and constrained decoding has done the same since 2024. The claim worth checking is calibration, and nobody has verified it.

    • Model evaluation
    • Structured output
    • AI agents

  2. 02

    Your prompt is telling Astra to do things it already does

    GPT-6 Astra runs tests, reads what it needs and reasons on its own. The instructions that used to help now cause over-testing and stalls. Which lines to delete, and the three dials worth setting instead.

    • Prompt engineering
    • Agents
    • GPT-6 Astra

  3. 03

    AI Agent Memory Engineering

    A deep-dive engineering guide to building memory systems for AI agents — short-term, working and long-term memory layers, with context engineering, RAG pipelines and persistent profile files.

    Research report

    • Agents
    • Memory
    • RAG

  4. 04

    From Classical ML to LLMs: Model Selection Guide

    A decision matrix for choosing the right model across 18 use cases — from tabular data to multimodal understanding. XGBoost vs LLMs, when to fine-tune, and what we would actually put in production.

    Research report

    • Model selection
    • LLMs
    • Machine learning