This portfolio is being built in public. Most sections are placeholder content and not yet functional — this is intentional, not a bug. Two case studies are live below: CS-001 — PCBO v2 · CS-002 — BEPO SDK
Designing architectures, methodologies
and AI-native systems
for complexity-heavy software domains.
Researching domain-driven design, operational architectures, context engineering and reproducible software systems. Building tooling where complexity demands more than convention.
Constraint-driven AI agent for hardware specification and validation. Combines LLM reasoning with a deterministic solver — physical constraints are a precondition of output, not a post-processing check.
Operational framework for designing reproducible software production systems. Formalizes the engineering process as a manufacturing model with defined inputs, transforms and quality gates.
Methodology for designing AI context as a first-class architectural artifact. Addresses context fragmentation across multi-domain knowledge workflows.
Reusable architectural templates for complexity-heavy domains. Each meta-model encodes a validated pattern: constraints, tradeoffs, operational model and known failure modes.
Designing a Constraint-Driven AI Agent for Hardware Specification Validation
→Built an AI agent that cannot produce a physically invalid hardware configuration. Core challenge: making constraint satisfaction a precondition of generation, not a post-processing filter. Validated against 200 physical assemblies at +32°C ambient.
A Typed Language for Business Models That Cannot Lie
→Built a TypeScript DSL where broken financial models fail at compile time, not at month six. Five-level Atomic Design architecture: each level validates its own invariants before the next composes from it. Evidence-bound hypotheses, unit economics as a compile gate, and a deterministic solver for exact breakeven thresholds.
Formalizing Software Production as a Reproducible Manufacturing System
Designing next. Operational framework for treating software development as a production system with defined inputs, quality gates, and measurable outputs.
Architectural pattern for building AI agents in constraint-heavy domains. Constraints are encoded as validation gates that block generation, not as post-processing checks. Forces the model to satisfy physical or logical reality before producing output. Replicable to any domain with hard rules and verifiable facts.
Operational framework that formalizes software development as a production system. Each development activity is modeled as a transform with defined inputs, quality gates, and measurable outputs. Eliminates ad-hoc process decisions and makes engineering work reproducible across teams and contexts.
Methodology for treating AI context as a first-class architectural artifact. Context is not a prompt — it is a structured knowledge object with boundaries, versioning, and ownership. Addresses fragmentation in multi-domain AI workflows by applying bounded context patterns from DDD to prompt and memory design.
Long-form engineering writeups. Architecture reviews. Methodology papers.
I design systems for domains where complexity is the primary constraint. My work sits at the intersection of software architecture, domain modeling, and AI-native engineering — building things where conventional patterns break down and the problem demands a framework before a solution.
Currently researching constraint-driven AI agent design, reproducible software production systems, and the application of systems engineering principles to knowledge-heavy domains.
I think in architectures. I document in frameworks. I build to prove the concept works — then formalize it so it can be replicated.