Principal AI Systems Architect | AI-native SDLC & Agentic Systems | Evals | Distributed Systems | Open Source Builder
I work at the intersection of AI systems, distributed software, cloud-native platforms, and practical software architecture. Today I architect the AI engineering transformation of a ~400-person, ~15-team business unit: prototypes first, then adoption by teams, measured against the historical baseline.
How teams move from AI-assisted coding to an AI-native SDLC: specs, context, evals and harnesses, tested in experiments rather than slides.
- The AI SDLC Manifesto: rethinking AI-native software development
- From Waterfall to Agile to Agentic AI: Software Engineering 3.0
- Why SDD is not a silver bullet, backed by the speckit_test experiment: why "the specification is the source of truth" conflicts with iteratively developed code
- The Software That Built Itself: well-defined intents are all you need
- Improve context before you improve your codebase, implemented as speckit_okf (a Spec Kit extension, listed in the community extensions) and catalogify (
pip install catalogify): Open Knowledge Format bundles built from code and git history and kept current as the code evolves
- Eval Driven Agentic AI Development: production-grade agentic AI with evaluation-driven development
- LLM-Trust-audit: black-box audits of LLM endpoints (change one detail that shouldn't matter, measure what changes), with paper, harness and run data
- Elevation Transformer: a cross-attention transformer surrogate for Irregular Terrain Model (ITM) path loss. ~32M samples on a single L4 GPU, RMSE 62.0 → 7.36 dB (median 2.99 dB); conformal lower bounds let a spectrum engine skip ITM calls that cannot change a grant. Paper
- Zarr elevation data system: USGS 3DEP terrain in Zarr chunked storage with LMDB tile keys and an R-tree: 10.6× compression (170 GB → 18 GB), 4.7–5.9M elevation lookups/s. White paper
- Tiny Transformer: a small language model built to explain transformers as simply as possible
- Data LakeHouse OS: Trino / Presto on Kubernetes with a standalone Hive metastore and Postgres
My work spans:
- AI systems, neural networks, and transformers
- Distributed systems and Software architecture
- Radio Network Management Systems,SON, PathLoss, Wifi6/7
- Data platforms and Developer infrastructure
- Cloud-native engineering (Kubernetes,gRPC,ArgoCD etc)
- Technical writing, papers, and Open source projects
- Transformer-Based Surrogate Model for Accelerated Irregular Terrain Model Path Loss Prediction
- High-Throughput Terrain Retrieval for Radio Propagation Modeling Using Zarr-Based Chunked Storage
- The Mathematical Intuition Behind Deep Learning: From the Dot Product to Multivariate Calculus and the Jacobian, with Python Amazon In Kindle Amazon Intl
- Explaining Transformers as Simply as Possible through a Small Language Model
- CNN from Scratch
- Vibe Coding — Prompts Are All You Need?
- MCP — The Golden Key for AI Automation
- REST is not the Best for Micro-Services — GRPC and Docker makes a compelling case
- Exploring Large Language Models — Language Understanding, Task Generalization & Fine-tuning
- Why Specification-Driven Development (SDD) is Not a Silver Bullet for AI-Assisted SDLC
- Training / Fine-tuning a Masked Language Model (MLM)
- Leveraging Smaller LLMs for Enhanced Retrieval-Augmented Generation (RAG)
- Backpropagation Deep Dive — Back Propagation with Softmax Output
- Micro Services & Git — Mono/Multi Repo, Balancing Feature Development
- End to End Testing in MicroServices
- From Data Warehouse to Data Lake to Data Lakehouse
- Using Apache Cassandra — A Few Things Before You Start
Context engineering for coding agents, evaluation-driven AI development, practical transformer systems, distributed architecture, cloud-native platforms, and production-grade engineering workflows.
Views expressed here are my own and do not reflect those of my employer.
Open source work on this profile is independent and does not include any proprietary or confidential employer material.