Writing
Thoughts on technology, AI, and building things.
Posts
- 2025Production-Grade RAG: A Blueprint for Scalable, Real-Time Architecture
- The Self-Correcting RAG: Implementing Agentic and Recursive Retrieval Loops
- We beat the benchmarks and ran out of runway
- Three clouds, one inference path
- From a ComfyUI graph to an API a brand's team can call
- Fitting a saree into 24GB: the context window trick
- FID was the wrong metric for the problem we had
- Composing adapters that disagree with each other
- Production Agent Architecture: Memory, Approval Gates, and Durable Checkpointing
- Architecting Multi-Agent Teams with LangGraph: Supervisor Patterns, Worker Isolation, and Sub-Graph Composition
- What happens when you merge two LoRAs
- Serverless GPUs for bursty training
- RAG Foundations: From First Query to a Retriever You Can Actually Trust
- One LoRA per failure mode
- Generic aesthetic scorers hate e-commerce photography
- LoRA finetuning: the hyperparameters settled in a week, the data never did
- Two streams: why one diffusion pass couldn't do virtual try-on
- What 5,000 images taught me about curation
- ComfyUI as a research lab
- Virtual try-on is not image generation