01
Problem
A fluent AI answer can sound trustworthy even when its inputs are stale, incomplete, contradictory, or weak.
Read the chapterIndependent project / AI & decision systems
SignalOS.
Rules decide.
AI explains.
You choose.
I built SignalOS to turn noisy stock and crypto research into a shortlist worth investigating. Rules surface candidates, AI explains the evidence, and I make the final judgment.
SignalOS / Demo Mode / sample data
My role
Design the judgment.
Then build the tool.
I designed and built the React/TypeScript prototype, using structured signals to surface candidates and AI to explain the evidence, not make the decision.
My contribution
- Solo Product Strategy
- UX & Systems Architecture
- Full-Stack Prototyping (React/TS)
- LLM Schema & Prompt Design
Active prototype
Demo Mode uses sample data and saves nothing.
The short version.
Good questions before quick answers.02
Approach
I reframed the work from chatbot prompting to decision architecture: rank attention, structure evidence, expose uncertainty, and preserve the user's judgment.
Read the chapter03
Solution
Deterministic rules set scores and constraints; AI explains the structured evidence; the interface keeps counter-evidence, portfolio context, and no-position states visible.
Read the chapter04
Outcome
The current demo proves the interaction model across radar, watchlist, opportunity review, portfolio context, risk rules, and a decision journal. It uses sample data and does not validate investment performance.
Read the chapterThe guided story
From AI answers to inspectable decisions
I started with a problem in my own research: too much information, and no clear way to decide what deserved attention. These decisions show how I separated useful signals from generated interpretation without letting the tool make the decision for me.
01PROBLEM
A confident answer is not enough
Consequential financial review depends on evidence quality, freshness, counter-evidence, timing, exposure, and the cost of being wrong. A persuasive generated thesis can conceal whether those conditions are weak or unresolved.
SignalOS / Demo Mode / sample data
02PROCESS / SYSTEM REFRAME
Map the decision before adding AI
I broke the process into stages: identify what deserves attention, organize evidence for and against action, form a provisional judgment, choose whether to act or wait, and preserve the reasoning for later review.
Interactive explanation / illustrative, not live scoring
A decision has a shape.
What deserves a closer look?
Meaningful changes enter a review queue. Being interesting is not the same as being actionable.
What this stage producesA shortlist worth investigating
An explanation of the design logic. No market data, scores, or decisions are generated here.
See this in the sample-data demo
SignalOS / Demo Mode / sample data
03DECISION / ATTENTION
Attention and quality answer different questions
A large change may deserve review without being a high-quality opportunity. The demo separates attention reasons from Trade Quality so urgency does not become a recommendation.
Interactive explanation / illustrative, not live scoring
Interesting ≠ ready.
Something changed.
Attention
It may deserve attention.
Readiness
Still unresolved.
A change can move an item up the review queue without demonstrating that action is justified.
An explanation of the design logic. No market data, scores, or decisions are generated here.
See this in the sample-data demo
SignalOS / Demo Mode / sample data
04DECISION / AUTHORITY BOUNDARY
Rules set the boundary; AI explains inside it
Deterministic rules produce the scores, thresholds, sizing constraints, and permissible states. The Analyst layer turns structured evidence into readable reasoning without setting the score itself.
Interactive explanation / illustrative, not live scoring
Rules. AI. You.
Set the boundaries.
- Produce scores and thresholds
- Define sizing constraints
- Constrain permissible states
↳ The generated explanation does not set the score.
An explanation of the design logic. No market data, scores, or decisions are generated here.
The boundary in the product.
Inspect the review and the rules behind it.
Sample PLTR opportunity review in Demo Mode. Deterministic Trade Quality and readiness remain distinct while the Analyst layer explains the current evidence.
Drag the frame. Pass it back.
Asset 1 of 205DECISION / TRACEABILITY
Keep the evidence and its weaknesses visible
The review exposes populated and missing evidence streams, source labels, dimension scores, risks, and confirmation requirements. The explanation remains connected to inspectable inputs.
SignalOS / Demo Mode / sample data
06DECISION / RESTRAINT
No position can be the responsible outcome
Portfolio exposure, risk budget, evidence quality, and user-defined boundaries can all prevent action. In the observed example, the product holds the opportunity at Watch and prefers Cash / No Position.
Interactive explanation / illustrative, not live scoring
Waiting is a real outcome.
Keep looking. Keep the question open.
A lead can remain worth reviewing while confirmation or readiness is unresolved.
An explanation of the design logic. No market data, scores, or decisions are generated here.
See this in the sample-data demo
SignalOS / Demo Mode / sample data
07CURRENT STATE / REVIEW
Preserve the reasoning for later review
The journal carries a decision beyond the moment of action by recording why it was taken or passed, what the plan was, what happened, and what was learned. The observed demo uses sample records and saves nothing.
SignalOS / Demo Mode / sample data
What the prototype demonstrates
A working decision architecture—not a validated investment product
The observed demo connects ranked attention, opportunity review, deterministic scoring, explanatory synthesis, source-level evidence, portfolio context, user risk boundaries, and a decision journal. It demonstrates the product model and its authority boundaries using sample data. It does not demonstrate investment performance, scoring accuracy, production readiness, or validated user outcomes.
EVALUATION STILL REQUIRED
The next stage is to verify the production feature set and technical architecture, test whether users understand the distinction between score and synthesis, evaluate evidence freshness and failure states, and define success around decision quality rather than financial outcome alone.
A better review, not a louder answer.
Explore the live prototype ↗Keep exploring
A different way to remember.
Memory Television — a personal archive explored through a handheld CRT.
Let’s make something
worth caring about.
Open to staff and lead product design roles, design engineering, and creative technology collaborations in Greater Boston and remotely.