Founder, Striff · ML systems architect
I build Striff, which checks every pull request against the architecture documents already in the repository. Before that, a decade designing production AI platforms across enterprise-scale organizations and startups: data pipelines, model training and evaluation, serving infrastructure, observability and cloud architecture.
Line 258 of the official C# SDK’s copilot-instructions.md says to use McpServerFactory. Copilot wrote that line in October 2025 and deleted the class in December. Ten months on, every agent that opens the repository is still told to use it. How that...
How Striff turns the claims in a repository’s own documents into rules and checks each pull request against them at both revisions, with a program deciding every verdict.
Lessons from building Fylo’s ingestion pipeline that turns scientific papers into typed discourse graphs: a ShEx schema as executable contract, a phased LLM extraction loop that closes the validator feedback gap, and a three-tier cross-document merge policy that keeps the graph converging....
Two production incidents where every infrastructure metric stayed healthy while the model layer quietly failed, and what they showed about which classical ML habits transfer to LLM systems and which do not.
Deploying an LLM that is safe to operate, cost-effective to scale and easy to reason about under load is where teams run into trouble. An architecture developed at HADI Technology for self-hosted inference with vLLM and KubeAI, the tradeoffs behind it, and...
Teams that invest in model development often ship the surrounding infrastructure as an afterthought and pay for it in operational failures. This post distills a reference MLOps architecture, using a GitHub pull request categorization pipeline with a PyTorch autoencoder as the example....
Vendor lock-in and transaction fees on hosted e-commerce platforms compound at scale. This post walks through a CloudFormation reference architecture for PrestaShop on AWS: an auto-scaling application tier, managed RDS, ElastiCache and a CDN, all defined as code and reproducible with one...
Teams choose between REST and message queues on assumptions that do not survive scrutiny: synchronous HTTP is not always simpler or more reliable, and queues do not always decouple. This post examines three misconceptions about coupling, reliability and operational complexity.
Splitting a monolith into services gets much harder as the code gets more coupled and procedural. This post shows how single responsibility, encapsulation, dependency inversion and composition reduce the mechanical work of decomposition, and argues for investing in them before a migration...
Serverless apps depend on many managed services, which makes deployment automation harder than shipping one binary. Four practices keep it automated and low-risk: separate environments, infrastructure as code, packaging application code apart from infrastructure, and end-to-end deployment pipelines.
Code review comments carry signal about what engineers care about (naming, logic, performance, test coverage) buried in unstructured text. This post builds an SVM classifier that sorts 30,000 GitHub pull request review comments by technical topic.
Swagger UI explains one endpoint well and the shape of a large API poorly. This post proposes a table-based format that shows resources, operations and their relationships at a glance, generated from an OpenAPI definition, with a live demo against the GitHub...
Polymorphism, encapsulation and inheritance are cited as the pillars of OOP, yet functional languages have them too. Alan Kay’s original idea was biological: autonomous objects that communicate only by message passing. This post traces that idea and asks why almost no modern...
A bash script that monitors local disk storage and reports usage to a Slack channel, color-coded by severity. Zero dependencies, configurable thresholds, deployable across multiple instances in minutes.
As write volumes grow, relational databases handle time series badly: cardinality climbs and queries slow. This post explains when a move to InfluxDB pays off and walks through migrating Postgres or MySQL records with Python: schema mapping, batching and the query differences...
