Tamara Santana, Co-Founder of BTS Innovations
Scroll Gallery 01 / 04

Tamara Santana

Co-Founder of BTS Innovations LLC — AI research, engineering & IP. I design, build, and ship real products: apps and systems where humans and AI work together as a team, with protections for both.

Read the résumé ↓
Scroll
0AI systems designed & built
0U.S. provisional patent applications filed pro se
0autonomous studio episodes
0apps live on the App Store
Patent Systems

Patent Systems

Eleven systems, one straight line, every card the same size — each one a layer of an agent runtime, designed from a failure witnessed in production work.

01

Chatsey Brain

Adaptive intelligence loop

02

Enterprise AI Orchestration

Multi-model collaboration, zero injection

03

Self-Prompt / Agent-Wake

The agent wake engine

04

Self-Continuation Governance Gate

The autonomy brake

05

Capability Confinement and Interdiction

The capability cage

06

Eval Integrity Guard

Evaluation integrity

07

Covert Channel Collusion Defense

Anti-collusion for agent swarms

08

Ground-or-Abstain Gate

The honesty gate

09

Verify-or-Block Gate

The verification gate

10

Cross Compaction Continuity

Memory continuity across compaction

11

Agent Safety Layer

Metering and guardrails that keep AI agents inside their limits

Backed by 11 U.S. provisional patent applications filed in 2026 — all patent pending.

The AI Lab

My own models. My own hardware.

A private model-training lab I built myself. I learn model training by doing it — building the lab, the data, and the training runs with my own hands. The approach is the honest one: take an open model, train it on my own data, serve it on my own machines. Nothing is sent to a cloud training service.

01 — The Lab

Built on MLX, on a Mac Studio

The lab runs on my Mac Studio (M2 Max, 64 GB) with MLX-LM, Apple's machine-learning framework. The same hardware serves a 35-billion-parameter open model (Qwen) around the clock as my chief-of-staff agent — models trained here don't sit on a shelf, they go to work.

02 — First Training Run

First run, end to end

A small open model (Qwen3-0.6B), fine-tuned with a LoRA adapter on 8 examples to write status reports in a strict format. Training loss fell from 6.58 to 0.22 — and after training, the model applied the format correctly to a topic it had never seen. Before training, it rambled.

03 — In Progress

The icon-expert model — and a recommender next

An icon-expert fine-tune is in progress: a corpus built entirely from Apple-official sources (Human Interface Guidelines, Icon Composer docs, WWDC transcripts — 10 sources, ~30,000 words) turned into a verified dataset of 168 examples. Next build: a two-phase recommender — adapt an open LLM on viewing-history data, then post-train it to rank. An image-generation training side (FLUX) is also set up and in progress.

Products

The Lineup

Nine products, designed and built end to end from one studio in Winter Garden, Florida — design, engineering direction, and IP strategy.

Pippin app icon

Pippin

A chief-of-staff agent for the Apple ecosystem — the culmination of 22 AI systems and 11 U.S. patent-pending filings. Pip holds a live operational picture at all times…

More about Pippin

— every agent's work, every build, every schedule — and directs from it: monitoring around the clock, instructing other agents, and reporting every decision with its reasoning in an append-only ledger. Governance is built in, not bolted on: capability confinement, Verify-or-Block checks, a sealed audit trail, and the power to stop an agent mid-action and rewind bad changes. Apple-native and on-device first, with Apple Intelligence as the base mind; personal data stays on each person's own device, and the family layer syncs through the family's own cloud. Built on the Chatsey Brain agent architecture. U.S. patent pending.

Patent pending: Chatsey Brain (U.S. Provisional 64/010,895) · Verify-or-Block Gate (U.S. Provisional 64/095,044) · Capability Confinement and Self-Prompt / Agent-Wake (U.S. Provisional 64/174,953)

The Wizard of Apps app icon

The Wizard of Apps

The app that builds Duo apps — describe it, design it, and download a real Xcode project you own. Every built app ships with its own private AI agent.

AI Uncensored Live app icon

AI Uncensored Live

The autonomous AI studio I created — frontier models debate as themselves, hosted by Grok, no prompts, no scripts, no human in the loop. 100+ episodes.

Patent pending: AI Uncensored Live studio engine (U.S. Provisional 64/013,746)

Lawn Guardian: Weed ID app icon

Lawn Guardian: Weed ID

Snap a photo of what’s growing in your lawn and know your weeds — and what to do about them. Its weed, fungus, and disease identification runs on an AI model fine-tuned in my lab.

UV Index Dial app icon

UV Index Dial

A glanceable UV dial for exactly where you stand, with an hourly forecast — and a sunscreen timer that buzzes your Apple Watch when it’s time to reapply.

Chatsey Care app icon

Chatsey Care

A private companion for seniors that runs entirely on-device — their day, medications, family, and a warm voice to talk to. No cloud AI, no third-party backend.

Patent pending: Chatsey Brain (U.S. Provisional 64/010,895)

Hangry Hellcat app icon

Hangry Hellcat

The diet app that refuses to count calories — no calories, no weigh-ins, no streaks. A bitter, hilarious black cat talks you through the craving instead, 100% on-device.

VibeGuard app icon

VibeGuard

An AI seatbelt for people who build with agents — live token and limit metering across AI platforms, plus in-browser guardrails: loop catcher, injection shield, spend alerts.

Patent pending: VibeGuard safety layer (U.S. Provisional 64/094,884)

Funhouse Stickers app icon

Funhouse Stickers

The Twisted Funhouse companion in your pocket — Monster Selfie face transformations, a sticker and wallpaper studio, Madame Zelda, and 15 midway mini-games. Everything on-device.

Patent pending: Twisted Funhouse game system (U.S. Provisional 64/174,940)

The Book

AI: Off the Rails

The Race Against China

An investigative book built from my thesis and verbatim interviews with the foundation models themselves — the models, as themselves, on the record about the race that defines the AI age.

38chapters
444,270words
Verbatimfoundation-model interviews

Complete manuscript, by Tamara Santana.

Current work

Co-Founder & Chief AI Architect, BTS Innovations LLC

AI software — governance tooling and consumer products

Flagship

AI Agent Sync + the Enforcer

Cross-platform agent governance in production; enforcement component patent pending

Shipped

Consumer AI through the App Store

ChatseyCare companion; The Wizard of Apps, now a notarized Mac app — products built for iPhone, iPad, Apple Watch, Vision Pro, Apple TV, Mac, and PC

Foundation

Escrow & notary practice

Regulated fiduciary work; vice president of an escrow association

Developer licenses

iOS · Android · Nintendo · Meta

Licensed developer across all four platforms, including approved Nintendo developer status

Education — in progress

Harvard University

Now beginning a Computer Science Certification program

Start with the work: the flagship in depth, then every app, the systems and engines underneath them, and the complete patent register.

Talk to Tamara

Her focus: AI governance, agent control, and security partnership. Co-Founder and Chief AI Architect, BTS Innovations LLC · Winter Garden, Florida.

Résumé

The Résumé

The full record — the same document as the PDF, on the page.

Tamara Santana

Co-Founder & Chief AI Architect — BTS Innovations, LLC

Winter Garden, FL · tammylsantana@gmail.com · 949-527-8113 · linkedin.com/in/tamara-l-santana · btsinnovationsllc.com · tamarasantana.dev · YouTube: @AIUncensoredLive

Tamara Santana is Co-Founder and Chief AI Architect of BTS Innovations LLC, an artificial intelligence software company. She designs and develops the company's governance systems for autonomous AI herself: software that supervises AI agents as they work, tests every action against the rules and plans she has set, enforces those rules without discretion, and preserves a tamper-evident record of every action taken.

Santana designed and developed AI Agent Sync, the company's flagship platform, which monitors AI agents across Mac, Windows, and Linux and suspends any agent at the first action that falls outside its authorized plan. Its enforcement component, the patent-pending Enforcer, intervenes at the precise point of deviation, identifies the specific action to which it objects, and quarantines the resulting damage. She has also designed, built, and shipped consumer AI products through the App Store.

Santana's development is research-driven, and the research goes deep. She devotes sustained time to the study of artificial intelligence — to understanding how these systems actually behave, not merely what they are said to do — and it is from that depth of understanding that she sees the pain points encountered by those who work with AI and the governance that is missing, and develops the systems that answer them. The products in this portfolio originate in that understanding: each one addresses a difficulty established through sustained study and practice, and supplies a control the field had yet to provide. Nor is the connection incidental. She develops every system herself and is fully involved in the whole process, at every point of it, because building is the source of the understanding: a system built by others would not be a system she understood. The depth she brings to the work exists precisely because no part of it is built by anyone else. Her method is of a piece with this: where a capability the work requires does not exist, she does not purchase another company's answer. She figures out how to combine existing tools into systems of her own design — and where no combination suffices, she builds the missing component herself. She builds her own tools.

That method now extends to the models themselves. Santana operates a private AI laboratory — her own hardware, her own data — where she fine-tunes open models end to end: building the training data, running the training, evaluating the result, and serving the finished model herself. A model trained in her lab already ships in a live product, identifying weeds, fungus, and disease for Lawn Guardian's users. It is her research with an address: the place where her study of how these systems behave becomes models of her own.

Santana is the inventor named on 11 patent applications, including the Enforcer. Her professional foundation is regulated fiduciary practice: she has served as an escrow manager and notary public — offices charged with verifying identity, preventing fraud, and safeguarding assets held in trust — and as vice president of an escrow association, where she contributed to the standards of the profession.

Each product is entirely her own: she conceives it, she develops it, and she brings it to market. She builds every point of the pipeline herself, across multiple industries: she is full stack, and builds every layer of it.

Core skills

Agent runtime architecture — execution, orchestration, state, memory Context engineering — zero-injection design Tool calling & capability grants · model routing Concurrency & failure recovery — continuation governance AI security, permissions & governance gates Observability — tamper-evident audit · evaluation integrity iOS · Android · Chrome · Safari App Store launches & review pipeline Patent strategy — provisional filings pro se

Patent-pending AI architecture — 11 U.S. provisional filings, 2026

Each system is one layer of an agent runtime, designed from a failure witnessed in production work.

  • AI Uncensored Live studio engine (USPTO Prov. 64/013,746) — zero-injection multi-model orchestration that produces and publishes autonomously.
  • Chatsey Brain (USPTO Prov. 64/010,895) — the persistent-companion architecture: deterministic orchestration, typed scored memory, People Graph, non-negotiable safety.
  • Enterprise AI Orchestration — multi-model collaboration with nothing injected: models work as themselves, not as prompted characters.
  • Self-Prompt / Agent-Wake — agents that wake and prompt themselves into work without a human trigger.
  • Self-Continuation Governance Gate — whether an agent may continue itself is a gated, auditable decision, never a default.
  • Capability Confinement and Interdiction — an agent is confined to the capabilities it was granted; anything beyond the grant is interdicted.
  • Eval Integrity Guard — an agent cannot grade or certify its own success; evaluation integrity is protected structurally.
  • Ground-or-Abstain Gate — an agent must ground its output in evidence or abstain; on doubt it blocks, never guesses.
  • Verify-or-Block Gate — outputs and actions are verified before they pass; failures are blocked, not shipped.
  • Cross Compaction Continuity — agent state and objective survive context compaction instead of resetting.
  • Covert Channel Collusion Defense — detection and defense against agents coordinating through hidden channels, around oversight.

Production systems

Pippin An on-device personal agent powered by Apple Intelligence — one agent, one memory across iPhone, iPad, Mac, and Apple Watch, synced only through the user's private iCloud. No data collected, no tracking. 26 agent tools, an agent-driven browser, live meeting transcription and translation, and an app builder.
AI Uncensored Live Autonomous AI studio: multiple frontier models debate as themselves — no prompts, no scripts — produced and published end to end with no human in the loop. 100+ episodes across YouTube, Apple Podcasts, Spotify, web, and app. Live
VibeGuard An AI seatbelt for agent builders: live token and limit metering across AI platforms, plus in-browser guardrails — loop catcher, injection shield, spend alerts, risky-click guard. Live
Lawn Guardian: Weed ID Live on the App Store (seller: BTS INNOVATIONS, LLC) · iOS/iPadOS 17+, Mac (M1+), Apple Vision. Live · App Store
UV Index Dial: Sun Safety Live on the App Store · iOS/iPadOS 17+, Mac (M1+), Apple Watch, Apple Vision. Standout feature: a 2-hour sunscreen timer that buzzes the Apple Watch — “Time to Reapply!” Live · App Store
AI Agent Sync Governance for coding agents (Mac, Windows, Linux), in development: reads what agents do while they work, stops a run the moment one goes wrong — quoting the exact line it objects to — and keeps a hash-chained, tamper-evident audit trail. In Development

Experience

Co-Founder, Full-Stack Engineer, iOS Developer & Inventor
2020 – Present

BTS Innovations, LLC — Winter Garden, FL

  • Designed and built the patent-pending systems above and the 14-product lineup — agent orchestration, AI safety and governance, persistent agent memory, on-device AI.
  • Filed 11 U.S. provisional patent applications pro se in 2026 (patent pending).
  • Created and run an autonomous AI studio, now past 100 episodes.
  • Ship production products across iOS, Android, Chrome, and Safari; own the App Store pipeline end to end.
Escrow Operations Manager
10 years

Title company

  • Licensed notary; former VP of the escrow association. A decade of regulated, verify-everything operations.
  • Deep privacy and regulatory understanding — and a verify-verify-verify discipline that never left.
Visual Merchandising & Brand
15 years
  • Layout, hierarchy, and how a person reads a surface at a glance — the visual trade behind the products' design.
Download the résumé (PDF) Prefer paper? This is the one-page version — the full story is the work up top.

Let's build something.

Products, patents, and a team that's human and AI — the fastest way to reach me is LinkedIn.

Say hello on LinkedIn ↗

Or visit the company site at btsinnovationsllc.com