Josiah Bujanda, Applied AI Systems

Summary

I came to AI systems through operating problems. Ten years estimating and delivering commercial construction taught me to care less about whether a model can produce an impressive answer and more about whether the whole system can be trusted to do useful work again tomorrow.

Today I design those systems end to end: workflow discovery, product definition, agent architecture, tool execution, state, evaluation, safeguards, production iteration. I own the design, the decomposition, the acceptance criteria, the debugging, and the production behavior. AI does a lot of the typing.

FutureKind, 2022 to present. Founder

Independent practice designing and operating agentic AI systems from discovery through deployment, evaluation, and iteration. The practice has its own page: FutureKind.

Stem, owned end to end, core bet that sensitive work should remain private without giving up access to frontier AI. Strategy, users, roadmap, prioritization, interaction design, evaluation, and production iteration are mine. stem

Stem's reliability loop. Incidents become permanent regression tests, privacy-aware routing that fails local by default, human approval gates for consequential actions, a 96/96 self-test gate, and degeneration detection calibrated against 30,230 stored turns. the Stem page

The bid-review product. Embedded discovery and workflow mapping through requirements, prototyping, evaluation, launch, and iteration. $221,915.70 surfaced across $3,442,234.02 of live bids, every finding linked to source evidence and held for human approval before it reached a customer. the review desk

Orchestration redesigned around product performance rather than maximum autonomy, dependency-aware parallel tasks under controlled execution, 2.25× realized end-to-end throughput. the orchestrator page

An internal quote import tool, shipped through beta into production for a 12-person estimating team, estimated to return 30-60 hours a week of manual re-entry; 12 releases, no rollbacks. the quote import tool

Thirteen working products across iOS and the web, primarily creative audio tools, two of them in Apple distribution, including Perihelion, a standalone and plugin synthesizer. the thirteen products

Horton Controls Group, 2015 to present. Senior Estimator and Project Manager

Preconstruction and project delivery across complex, high-value commercial developments, with responsibility spanning technical scope, stakeholder coordination, schedule, commercial risk, and field execution.

Developments led include Globe Life Field, Dickies Arena, Goldman Sachs’ Dallas headquarters, Liberty Mutual, Legacy West, Texas Live!, Trammell Crow Center, and Pegasus Place.

Represented the controls scope in stakeholder meetings with owners, developers, general contractors, engineers, manufacturers, and field teams, including direct coordination with major development organizations on projects facing severe liquidated-damages pressure for delayed delivery.

Owned execution across cross-company dependencies: changing requirements, technical constraints, procurement schedules, field conditions, and customer commitments, translated into decisions that kept work moving toward fixed milestones.

Managed scope across design interpretation, submittals, procurement, construction, commissioning, and closeout.

Across 1,600+ estimates, 300+ submittal packages, and 200+ commercial projects, including responsibility for lighting-controls scope on projects totaling $250M+ in construction value and contributions to developments totaling $2.5B+ in publicly reported cost.

Product and technical

• Product management

  0-to-1 product development, product strategy, customer discovery, problem framing, product requirements, roadmap development, prioritization, rapid prototyping, UX and interaction design, experimentation, success criteria, launch, product evaluation, iteration, human-in-the-loop design, stakeholder alignment.

• AI products and systems

  Agentic workflows, multi-agent orchestration, tool execution under permission gates, persistent execution state and resumable runs, context assembly, model routing, tracing, RAG, retrieval and natural-language search, AI evaluations, regression testing, deterministic validation, privacy-aware AI, computer vision, human approval systems, OpenAI and Anthropic APIs.

• Data and development

  Python, SQL / SQLite, JavaScript / React, APIs, Git, Jupyter, structured data models, taxonomies and classification systems; Claude Code in the implementation loop.

Education

B.S. Mechanical and Energy Engineering, University of North Texas, 2015.

The Premise

I build things to understand them. This is where I publish the work before it gets compressed into a résumé line, a case study, or a product category.

I build AI systems, browser games with my kids, software instruments, research pipelines, workflow tools, and whatever else a question seems to require.

The projects look unrelated. I am studying the same thing in all of them: how understanding forms, how it breaks, and what a tool can do about it.

The bid-review system and the roller coaster toy come from the same place. I have always been a builder; I started with Legos and I cannot stop. Building is how I look harder at how something works.

Method: build the smallest real thing that can push back on the idea. Bias: every number I publish traces to a receipt, and the person accountable for the outcome makes the call. The commercial work is under Work; music software lives at DeepField Studio.

Contact

El Paso, TX · hello [at] josiahbujanda [dot] com ·