My personal corner of the web — writing, open-source projects, agents, datasets, and talks from two decades of building data platforms. True to form, the whole site is data too: every list here is a parquet file.

Over the past few days, I have been exploring a connected problem. AI engineering practices are evolving rapidly, but estimation, pricing, productivity measurement, and value realization are still catching up.

AI engineering is moving very fast. Every week, we hear about new models, new agent development frameworks, Vector databases, RAG pipelines, MCP servers, Connectors & Tools, evaluation frameworks, and automation platforms.

Across decades of working on large enterprise products, multiple projects, and top-tier engineering teams, I have rarely seen a complete and dependable solution for two persistent problems: Impact analysis and bidirectional traceability. This is especially difficult for enterprise products that have

This is a sequel to Last week's post on the Claimwise dataset and dbt pipeline. Last week I shared the Claimwise dataset and the dbt pipeline behind it.

Production-Grade Agents, Built in the Open — Part 3 of 6 One Core, Four Doors Specs, Not Vibes Anatomy of a Bolt (this article) The LLM Never Scores Stop Burning Tokens on Grep AI Did the Typing Part 2 gave you the totals: 11 Bolts, 62 tasks, 135 evals. This part picks one real Bolt — the cover-lett

Claimwise is a healthcare Revenue Cycle Management (RCM) demo I built — the billing back office of a fictitious hospital network. The hospital is not real, but everything running behind it is real: synthetic patients and claims, all of it sits in one repo, and you can run it on your laptop.

Dashboard: senthilsweb.github.

Production-Grade Agents, Built in the Open — Part 2 of 6 One Core, Four Doors Specs, Not Vibes (this article) Anatomy of a Bolt The LLM Never Scores Stop Burning Tokens on Grep AI Did the Typing The agent from Part 1 was built through 62 tasks and 135 evals across 11 Bolts — every one of them writte

Today, I am sharing my homelab infrastructure—my small personal data centre where many of my ideas and projects have taken shape. After 26 years in technology, my work naturally spans architecture, engineering leadership, and management, but I have never wanted to move away from hands-on engineering

The titles AI Engineer and Forward-Deployed Engineer may sound new, but many of the qualities behind them are not. In my view, an elegant software engineer may already possess nearly 70–75% of the foundation needed for these roles.
Building a Generative Q&A Engine with Semantic Search and Embeddings
In the age of AI-powered tools and intelligent systems, the ability to search, analyze, and generate insights from documents has become paramount.
Edge-Optimized GenAI Chat Web-UI for Ollama & Cloudflare AI models with Context Injection
It is a simple GenAI chat web interface built with Next.js that connects to both local Ollama and Cloudflare AI models.
Embedding LLM in SQL is simple!
Inspired by Snowflake’s AI-driven analytics and MotherDuck’s recent implementation of SQL + LLM (Introducing the prompt() function), I explored LLM-powered Business Insights using DuckDB, embedding AI directly into SQL workflows for real-time data intelligence.
Data Pipeline Diagrams as Code
Creating diagrams for data pipelines, workflows, and architectures is a routine task for many of us. Whether it's for internal presentations, client discussions, or technical documentation, the process can often feel repetitive and time-consuming.
Privacy Engineering with LLMs: Safeguarding PII While Enhancing Text with Privacy Shield
In the age of AI-powered tools and intelligent systems, the ability to search, analyze, and generate insights from documents has become paramount.
Lightweight HTML to PDF with Puppeteer: Simple PDF Export for Edge Deployments
This microservice is designed to address simple PDF export needs, making it ideal for generating blog articles, profile cards, and other lightweight use cases.
Monorepo of Vercel Eve agents (API test generator, diagram generator, GitHub PR digest, LinkedIn cover generator) sharing a common runtime kit, plus the AI SDET Workbench slide deck.
largeOSSAIAgentsTypeScriptLive demoHealthcare RCM synthetic data generator and data lakehouse demo monorepo — inspired by dbt's jaffle_shop.
largeOSSHealthcareData-EngineeringdbtLightweight developer platform with Lua serverless functions, SQL playgrounds, embedded BoltDB storage, and AI chat assistant. Config-driven, single-binary deployment.
largeOSSGoPlatformCMSLive demoCollection of data engineering projects using Dagster to manage and orchestrate data pipelines.
largeOSSPythonDagsterData-Engineering