Follow your curiosity.

Mamba vs ViT at the Edge: A Post-Mortem

Why we ripped out our Vision Transformers for State Space Models in production.

When QLoRA Fails You

A deep dive into rank deficiency and catastrophic forgetting when fine-tuning Llama-3 70B.

Taming PPO for High-Frequency Execution

Notes on reward clipping and advantage estimation in noisy financial environments.

Agentic RAG: Moving Beyond Naive Retrieval

Why standard semantic search is dead, and how agents are taking over retrieval.

Scaling vLLM with Triton Inference Server

Achieving 5k tokens/sec on an A100 cluster.

Deep Learning on Tabular Data is Still a Meme

A retrospective on 10 Kaggle competitions and why XGBoost still reigns supreme.

Designing a Multimodal Coding Agent

Feeding screenshots and DOM trees into an autonomous web-testing agent.

Distilling Intelligence via Synthetic Data

Using frontier models to generate high-quality Kaggle datasets for small model tuning.

Fighting Concept Drift in Production

Implementing an online learning pipeline for our recommendation engine.

Scaling Test-Time Compute: The New Paradigm

Why giving your agent more time to "think" beats training a larger model.

Diffusion Models for Time Series Forecasting

Replacing autoregressive models with score-based generative models for market prediction.

Multi-Node Training Nightmares

Debugging NCCL timeouts with Hugging Face Accelerate and FSDP.

Zero-Shot Anomaly Detection using CLIP

Using Vision-Language models to detect manufacturing defects without labeled data.

LoRA vs. Prompt Tuning vs. Adapters

Benchmarking Parameter-Efficient Fine-Tuning (PEFT) methods.

The Routing Collapse Problem in MoE

How to prevent your Mixture of Experts model from only using two experts.

Feature Selection in High-Noise Environments

Why adding more features usually destroys your trading model.

Running Transformers in the Browser via WebAssembly

Client-side ML is finally viable for real workloads.

Designing Long-Term Memory for Agents

Moving beyond simple sliding windows for LLM context.

Detecting Fraud with Graph Neural Networks

Modeling transaction networks to catch bad actors.

Writing Custom CUDA Kernels for Attention

When standard PyTorch operations aren't fast enough.

RL / FIELD NOTES

The Sim2Real Gap in Robotics RL

Why policies trained in simulation fail catastrophically in the real world.

You Are Evaluating Your LLMs Wrong

Why human-in-the-loop and LLM-as-a-Judge are mandatory.

Deep Dive: Order Book Imbalance Signals

Extracting microstructural alpha from Level 2 data.

VLMs for Autonomous Navigation

Combining visual inputs with semantic reasoning for path planning.

Securing Agentic Workflows Against Prompt Injection

How we sandboxed our code-executing agents.

Federated Learning for Medical Imaging

Training deep neural networks without centralizing sensitive patient data.

Squeezing 70B Performance into 8B Weights

Advanced knowledge distillation techniques for Large Language Models.

Parsing SEC Filings with FinBERT

Extracting sentiment alpha from 10-K and 10-Q reports.

Tuning HNSW Indexes for Vector Search

Balancing recall and latency in billion-scale vector databases.

Autogen vs LangGraph: A Developer Perspective

Choosing the right orchestration framework for multi-agent systems.

Self-Supervised Vision with SimCLR

Learning visual representations without a single labeled image.

DPO: The Death of RLHF?

Why Direct Preference Optimization is taking over model alignment.

Hidden Markov Models for Market Regimes

Making models aware of changing market volatility states.

Maxing Out GPU Utilization with TensorRT-LLM

Compiling Transformers for Nvidia hardware.

Building Resilient Scraping Agents

Using Playwright and VLMs to navigate dynamic single-page applications.

PatchTST: Transformers that Actually Work for Time Series

Solving the noise problem in financial transformer models.

Training Vision Models on Unreal Engine Data

Closing the reality gap with domain randomization.

Reducing API Costs with Semantic Caching

Don't pay OpenAI twice for the same question.

RL / FIELD NOTES

Curriculum Learning for Complex RL Tasks

You can't teach a model to run before it knows how to walk.

Differentiable Portfolio Optimization

Integrating Markowitz Portfolio Theory directly into neural networks.

Multimodal RAG: Processing Charts and Tables in PDFs

Text-extraction ruins PDF structure. Vision models fix it.