Collinear AI’s Blog

RL Infrastructure for AI Agents: Why Environment-as-a-Service is the Missing Piece

Reinforcement learning for large language models is more of a systems problem than ML.

Nov 18 • 

Nazneen Rajani

3

1

Announcing Spider: a lightweight tool to craft post-training data recipes

TL;DR Spider is a single client interface that turns messy distillation and ablation experiments into a simple, configurable workflow.

Nov 6 • 

Soumyadeep Bakshi

6

Collinear Newsletter #8 – Notes on Improving AI

Hi AI innovators,

Nov 4 • 

Soumyadeep Bakshi

8

October 2025

The case for simulations

Unlocking model uplift through better evaluations

Oct 23 • 

Soumyadeep Bakshi

 and 

Grant Griffith

5

Through the Valley of Reasoning: What Small Models Teach Us About Learning

NeurIPS paper on knowledge distillation scaling laws for small foundation models

Oct 9 • 

Muyu

, 

Tsach Mackey

, 

Anand Kumar

, 

Meghana A Rajeev

, and 

Nazneen Rajani

Introducing Collinear Simulations: Steerable Personas for AI Agent Testing

TraitBasis, inspired from mech intrep, gives high-fidelity user personas for comprehensive agent testing

Oct 7 • 

Meghana A Rajeev

, 

Tsach Mackey

, 

Muyu

, 

Anand Kumar

, and 

Nazneen Rajani

2

Collinear Newsletter #7 | Notes on Improving AI

A deep dive into how leading companies are deploying AI agents reliably in production

Oct 2 • 

Soumyadeep Bakshi

 and 

Nazneen Rajani

5

September 2025

Introducing Curator Evals: A Benchmark for High-quality Post-training Data Curation

High-quality datasets are the foundation of better language models.

Sep 2 • 

Muhammad Ali Shafique

, 

Tsach Mackey

, 

Meghana A Rajeev

, 

Soumyadeep Bakshi

, 

Muyu

, 

Anand Kumar

, and 

Nazneen Rajani

2

August 2025

Collinear AI Now Available on Google Cloud Marketplace

Making safe, high-performing AI accessible through trusted enterprise infrastructure

Aug 25 • 

Soumyadeep Bakshi

7

You Can’t Hire Your Way to Model Alignment

Why the Global AI talent shortage Is undermining enterprise model alignment, and what you can do instead

Aug 13 • 

Marc Moring

, 

Soumyadeep Bakshi

, and 

Nazneen Rajani

2

Leveling the Playing Field: Livecodebench’s Big Bug Fix

Three major fixes that reshaped competitive coding scores and why your numbers may look very different now

Aug 12 • 

Muyu

, 

Tsach Mackey

, 

Muhammad Ali Shafique

, 

Anand Kumar

, 

Meghana A Rajeev

, and 

Nazneen Rajani

5

OpenAI's gpt-oss on LiveCodeBench: A Competitive Programming Deep Dive

tl;dr: the gpt-oss-20b is a strong model for competitive coding but is >3x sample inefficient compared to deespseek-r1-0528

Aug 6 • 

Muyu

, 

Tsach Mackey

, 

Anand Kumar

, 

Meghana A Rajeev

, and 

Nazneen Rajani

2

3

1