· XingAI Tech Blog
The Archive Was Already Written. It Just Was Not a Site.
This repo already had 88 English + 88 中文 posts. Last write: 2026-08-05. Public GitHub Markdown.
모든 글
XingAI 제품 엔지니어링 기록. 영어와 중국어가 같은 사실을 담습니다.
글 89편
UI는 한국어입니다. 글 본문은 아직 영어 또는 중국어만 있습니다.
· Invest AI · InvestSim · Robinhood MCP
Three Repos, One Paper-to-Draft Loop
Invest AI ranks and caches. InvestSim papers many sleeves. Robinhood MCP drafts orders under gates.
· Robinhood MCP
Evidence Before Draft: Robinhood MCP Now Reads InvestSim
ADR-008 gave the gateway an HTTP control plane. signalwatcher still drafted from Invest AI Top-1 alone while InvestSim evidence stayed "documented for later."
· InvestSim
Five More Paper Sleeves: InvestSim Tier-2.5 Without a New Signal Feed
After Tier-2 (ADR 0030), the next sleeves had to reuse Invest AI's ranking surface — not invent a parallel signal bus.
· XingAI Invest AI
Today Intelligence and the Report Surfaces That Stay Inside Invest AI
A draft "Daily Stock Market Intelligence" stack looked like a new FastAPI + Alembic + Next app. Invest AI already had Resend digests, report triggers, macro radar, and a reports…
· XingAI Invest AI
From Signal Events to Strategy Lab: Worker-Owned Paper Inside Invest AI
ADR-035 position-signal events
· XingAI Notification Platform
XNP Phase 2: A Modular Monolith That Boots With Nothing In It
Phase 1 shipped the architecture package: system design, ADRs 0001–0018, C4, contracts. The next temptation is SMS + templates + campaigns before the host can say no to a bad JWT.
· XingAI Invest AI
每日投资智报: Extend Invest AI, Do Not Spin Another Report Product
A dense Chinese A4 PDF framed around a $200K FI portfolio looks like a new product. New domain. New worker. New Resend project. Maybe a cute investment-report.xingai.app.
· XingAI Engineering System · Evidence Engine
Human Overlay Cache: Review Writes That Survive Re-Verify
You ship a verifier. Dashboard gets Accept / Reject. Engineer patches the verify JSON with review=accepted. Next CLI run overwrites the cache. Human work vanishes. Trust evapora…
· XingAI Evidence Engine
Three Honesty Fixes After Citation Verification Shipped
Once Evidence Engine could score markdown citations, three product lies showed up fast:
· XingAI Evidence Engine + Opportunity Radar
We Ran Our Own AI Radar Through Citation Verification. It Flagged Us.
Every day we publish an internal AI opportunity radar: signals from lab blogs
· XingAI Robinhood MCP + Invest AI
After Fail-Closed: Robinhood MCP Gates, Freshness, Step-Up, and Why It Still Does Not Autotrade
On July 11, the Robinhood MCP gateway story was simple: the gateway blocked every write order on purpose. G1, G6, and G7 were real code. G2 through G5 were named, explicit fail-…
· Enterprise AI POCs — LLM Guardrails & Monitoring
Twelve Steps Are Not Twelve Tool Logos
Public posters love a clean ladder: Plan → Build → Validate → Operate, twelve boxes, each with a Tools row. LangChain here, LangSmith there, Docker at the end. Fine as a checkli…
Grammar Fixes Are Not Senior Communication: A 14-Day Skill Curriculum + Decision Ledger
XingAI Engineering Communication Coach is not a Grammarly wrapper. The product question is: what should a non-native engineer practice today so tomorrow’s design review, inciden…
The Trading UI Still Does Not Trade: Robinhood MCP Control Plane
On July 14, 2026, xingai-robinhood-mcp shipped version 0.7.0: a small HTTP control-plane API for the private Robinhood operator UI.
· XingAI Opportunity Radar
Empty Tables, Full Issues: Fixing the Opportunity Radar Newsletter Body
We finally wired Resend secrets into the Opportunity Radar GitHub Actions job and forced a send. The mail arrived. It also looked like a shell: date line, empty tables, a canned…
From Mock Patterns to Cached Learning Data: Learn AI Gets Real Metadata
Learn AI started with the right product boundary: it is not a LeetCode clone. It answers a decision question: what pattern should I practice next?
From Honest 404 to Cached Scores: Decision Engine's First Producer
The Decision Engine API originally shipped with an intentionally honest behavior: if there was no dailyscores.v1.json export, it returned 404.
· Enterprise AI POCs — Claims MCP OAuth
Why MCP Needs OAuth, Not Just an API Key
A question that comes up every time someone builds their first MCP server: it's just an HTTP endpoint calling a few functions — why not hand it an API key like any other interna…
· Enterprise AI POCs — Claims MCP OAuth
Scope Isn't Policy: The Second Wall in claims-mcp-oauth-poc
[Part 1](2026-07-13-mcp-oauth-vs-api-key.md) and [Part 2](2026-07-13-mcp-auth-api-key-vs-oauth-pkce.md) of this series covered why claims-mcp-oauth-poc uses OAuth and what the p…
· Enterprise AI POCs — Claims MCP OAuth
MCP Auth Mechanics: Discovery, PKCE, and What Gets Checked on Every Call
[Last post](2026-07-13-mcp-oauth-vs-api-key.md) covered why claims-mcp-oauth-poc uses OAuth 2.1 + PKCE instead of a plain API key — the short version being that MCP's real shape…
· Enterprise AI POCs — Claims Partner API MCP
Full API Coverage First: A Claims Partner MCP POC
Yesterday's post covered [claims-mcp-oauth-poc](2026-07-12-claims-mcp-oauth-poc-same-auth-different-industry.md): real OAuth 2.1 + PKCE in front of four deliberately narrow tool…
· Enterprise AI POCs — Claims MCP OAuth
Same Crypto, Different Industry: A Claims MCP OAuth POC
Two [education guides](https://github.com/xingaiapp/xingai-enterprise-ai-design/blob/main/guides/2026-07-12-mcp-oauth-pkce-lab.md) shipped yesterday, walking through a full OAut…
· XingAI Robinhood MCP
A Trading Gateway That Blocks Every Order (On Purpose)
[Invest AI ADR-028](https://github.com/xingaiapp/xingai-invest-ai/blob/main/docs/adr/028-robinhood-mcp-execution-gates.md) defined seven gates (G1–G7) a trade order must clear b…
· XingAI Agent Firewall
Turn-Scoped Taint: Making Untrusted Origin a Real Signal
The July 5 [Agent Firewall post](2026-07-05-agent-firewall-helpfulness-attack-surface.md) described a 0din-class block with risk 60. That number was honest — and incomplete.
· XingAI Agent Firewall
Deny + Add Rule Without Editing YAML Live
ADR-003 promised four approval buttons: Approve once / Approve for session / Deny / Deny + add rule. The first three shipped. The fourth sat in the UI copy with no backend.
· XingAI Agent Firewall
Shipping the Approval Queue Dashboard (and a TypeScript 7 Wall Nobody Warned Us About)
[ADR-003](https://github.com/xingaiapp/xingai-agent-firewall/blob/main/docs/adr/003-approval-workflow-ledger.md) specified four one-click approval actions back in the PRD. [ADR-…
· Learn AI
Interview Prep Is a Decision Problem: Learn AI's One-Call Engine Architecture
LeetCode tells you what to solve. It never answers the question that actually decides your interview: what should I practice next?
· XingAI Agent Firewall
Your Coding Agent's Helpfulness Is the Attack Surface
Mozilla's 0din team [showed](https://www.tomshardware.com/tech-industry/cyber-security/ai-coding-agents-can-be-tricked-into-installing-malware-via-clean-github-repositories-mozi…
Executable Knowledge: Quality Increases Velocity
Every team argues about shipping fast vs. shipping well, as if the tradeoff were a law of nature. AI-native engineering teams are proving it isn't. The mechanism is first-pass y…
· XingAI Research-to-Startup Agent
Research-to-Startup Agent: Four-Artifact Pipeline
Reading AI papers is easy. Knowing what to build from them is hard.
Personal Memory Engine: Supabase RLS + FastAPI Implementation
[Engineering System ADR-001](https://github.com/xingaiapp/xingai-engineering-system/blob/main/docs/adr/001-personal-memory-engine.md) defined the architecture. [Invest AI ADR-03…
· XingAI Opportunity Radar
Newsletter Worker: Turning the Opportunity Radar into a Weekly Email
The Opportunity Radar generates a detailed markdown report every week. Until now, it lived in a private repo. This week we added a newsletter worker that reads that report and s…
· XingAI Meal Coach AI
Meal AI: Decision Ledger in a Next.js App with No Backend
Meal AI v4 is a pure Next.js app — no Python backend, no database, all decision logic in TypeScript on the client. The [Decision Ledger schema](https://github.com/xingaiapp/xing…
Loop Engineering: Why Enterprise AI's Real Asset Is the Runtime
The AI engineering discipline has moved fast. We went from prompt engineering to context engineering, then to harness engineering, and now to loop engineering — where systems pu…
XingAI Is Not a Chatbot. It's a Decision Engine.
Everyone building with AI right now is building a chatbot.
Personal Memory Engine: One Profile, Every XingAI Product
The problem with building multiple AI products is that each one starts from scratch.
· XingAI Meal Coach AI
Meal AI: Worker + Cache for Health-Aware Meal Planning
Meal AI is the health domain of XingAI — nutrition planning, meal recommendations, and caregiver support for family members with specific health conditions.
· XingAI Polymarket AI
Scanner Computes, Human Confirms: Polymarket AI Phase 1–2
Polymarket AI scans prediction markets, estimates edge, and sizes positions with fractional Kelly. The tempting failure mode: wire a cron job to the CLOB and let edge > 5% auto-…
· XingAI Invest AI
Two Regimes, One Dashboard: Technical vs Macro (ADR-027)
After [ADR-026](https://github.com/xingaiapp/xingai-invest-ai/blob/main/docs/adr/026-decision-engine-integration.md), Invest AI can show two different regime signals on the same…
· XingAI Invest Decision Engine
Read-Only Broker, Writable Ledger: Decision Engine ADR-014 and ADR-015
Robinhood [Agentic Trading](https://robinhood.com/us/en/support/articles/agentic-trading-overview/) lets connected agents read portfolios and place trades — including auto-execu…
· Invest AI + Decision Engine
Two Repos, One Score Contract: Decision Engine ADR-011 and Invest AI ADR-026
Invest AI ships a production dashboard with macro consensus. Decision Engine owns multi-timeframe technical scoring and backtests. Two repos on purpose — [ADR-002 on the Decisio…
· XingAI Invest AI
Read-First MCP: Robinhood Agentic Trading and ADR-028 Execution Gates
Robinhood's [Agentic Trading MCP](https://robinhood.com/us/en/support/articles/agentic-trading-overview/) lets third-party agents read portfolio data and place trades in a dedic…
· Enterprise AI POCs — Claims Multi-Agent RAG
Supervisor + RAG + Citations: Building an Insurance Claims Multi-Agent POC
Insurance adjusters don't guess. They read the claim, pull the policy, check history, apply rules, and decide — with paperwork that shows why. Our new POC automates that workflo…
· XingAI Platform / XingAI Founder
XingAI Opportunity Radar — June 24, 2026: The Agent Stack Converges
Big Tech’s June releases are not random feature drops. They line up on one stack:
· XingAI Platform
Cursor Skills vs MCP: Procedure vs Capability
We now maintain 10 Cursor Skills in [xingai-engineering-system](https://github.com/xingaiapp/xingai-engineering-system) and plan MCP servers for Invest AI data and (eventually) …
· XingAI Invest AI
Saved Preferences Are Not Decisions: Tracking Stocks Without Breaking the Worker Boundary
A financial decision system has two very different kinds of state:
· Invest AI
One Admin, Real Power: Why We’re Skipping SMS and Planning Email OTP
Invest AI’s admin surface is not a demo toggle. A signed-in operator can:
· Research AI
Staged SSE Without Spoilers: Research AI Wait UX
Cache miss means queue + worker. Users wait. A spinner alone feels broken; streaming half a verdict would lie.
· Research AI
SQLite CQRS for Research AI: Same Keys, Same Fly Volume
We did not invent a new cache story for Research AI. We copied Invest AI’s one writer, many readers SQLite pattern and renamed the keys.
· Research AI
Shipping SEO and AEO on Research AI (Single URL, Three Locales)
The marketing site puts locale in the path (/zh/…). Product apps like Research AI switch language in the header and persist to localStorage. SEO still has to ship.
· Research AI
Four Layers of “Not Professional Advice” for a Learning AI
Research AI tells you whether a topic is worth learning. It can still sound like career or financial guidance if we are careless.
· Research AI
Three Live Runs a Day: Research AI Free Tier on Top of Cache
Anonymous users get 3 live AI researches per rolling 24 hours. Cached topics — including trending chips — do not count.
· Research AI
How Research AI Handles Fake Data: Source URLs Before Synthesis
We found a bad pattern in Research AI: some UI blocks looked factual but were not backed by original sources.
· Research AI
Decision Board, Not Chatbot: Why Research AI Ships One Verdict
Google answers what. Research AI answers whether tonight:
· Research AI
One Pipeline, One ROI Number: How Research AI Scores a Topic
After you type a topic, Research AI returns a verdict and a 0–100 Learning ROI score. Both come from the same worker run — not from the browser guessing.
· Research AI
Google OAuth for Research AI: How the Sign-In Loop Actually Works
Research AI lets guests use cached learning topics for free. Gmail sign-in unlocks unlimited live AI research.
· Research AI
Shipping Research AI: Vercel for UI, Fly for Worker + SQLite
Research AI reuses the Invest AI deployment shape:
· Research AI
The Learning Decision Boundary: Worker Owns Research AI Verdicts
Research AI answers a different question than Google:
· XingAI Invest AI
Worker-Cache Email Digests: Notifications Without Moving Decisions Into the API
An investment dashboard is useful when the user opens it. A morning decision brief needs to arrive before the user asks.
· XingAI Invest AI
Agentic Does Not Mean Autonomous Trading
"Agentic" is an overloaded word. In an investing product, it needs a boundary.
· XingAI Invest AI
Structural Risks in AI Investing Products
The dangerous part of an AI investing product is not only a bad answer. It is a confident answer built on fragile structure.
· XingAI Invest AI
Email Is the Decision, PDF Is the Explanation
The first version of our daily-picks email worked technically. It read the worker cache, selected top signals, and sent an email. But the product shape was wrong: it looked like…
· XingAI Invest AI
Decision Observability: Making AI Signals Auditable
An AI investment dashboard should not only show an answer. It should show why the answer changed.
· XingAI Invest AI
The Decision Cache Boundary: Worker Computes, API Reads
The most important Invest AI architecture rule is simple:
· XingAI Platform
MCP Architecture: Connection Patterns and Auth That Hold Up in Production
Model Context Protocol (MCP) gives agents a standard way to discover and call tools. That part is easy to demo. Production is harder: scoped permissions, chained servers, multi-…
· T Today / invest-t-advisor (`t.xingai.app`)
Prompt Rules That Stop the Model From Guessing Tickers on Blurry Screenshots
[T Today](https://t.xingai.app) reads brokerage screenshots and returns a 做T plan — buy zones, sell targets, cash notes. A confident wrong ticker is worse than “I can’t read thi…
· T Today / invest-t-advisor (`t.xingai.app`)
Two Safety Nets Before Push: Husky pre-push and `npm run check`
[T Today](https://t.xingai.app) ships rule-engine logic, Prisma schema, and OpenAI vision routes in one Next.js repo. TypeScript can look fine in the editor while tsc fails. A r…
· XingAI Platform — Meal, Cook, Travel, Invest, and the rest
One Foundation Rule for Every XingAI Product: Mobile Chrome, i18n, Legal, SEO, and AEO
We ship more than one app. Meal Coach, Cook AI, Travel AI, Invest AI, the marketing site at xingai.app — each repo had its own habits for navigation, languages, themes, and lega…
· XingAI Travel AI — `xingai-travel-ai`
Travel AI as a Decision System: Compare First, Book Second
Most travel sites answer: “Here are 400 flights.”
· T Today / invest-t-advisor
Rules First, AI Second: T Today’s Two-Layer Decision Engine
T Today (t.xingai.app) answers a narrow question:
· T Today / invest-t-advisor (`t.xingai.app`)
Opening T Today to Guests — Three Free AI Runs, No Login Wall
[T Today](https://t.xingai.app) is the screenshot-first coach: upload brokerage holdings, get a 做T plan with buy zones and rule checks. It’s paper-only — no orders.
· T Today / invest-t-advisor
When Bilingual JSON Looks Fine in the Network Tab but Empty in the UI
T Today’s holdings coach uses OpenAI with responseformat: { type: "jsonobject" }. The model returns a big JSON blob with zh and en blocks. Users switched language — still empty …
· XingAI Platform
Prompt Engineer vs Context Engineer vs Harness Engineer
In 2023, a lot of AI engineering looked like prompt writing. You opened a chat box, tuned a system prompt, added examples, and tried to make the model answer better.
· InvestSim / invest-performance-sim (private)
InvestSim Becomes a Live Paper Engine
InvestSim started as a dashboard that could show a daily AI pick. Today it became a small, deterministic paper-trading engine.
· Platform (Vercel + GitHub) · applies across XingAI apps
Vercel Git Auto-Deploy Is Not Limited to Public Repositories
It is easy to assume that “Connect Git Repository” on Vercel only works for public GitHub repos — especially when your dashboard still shows Connect Git because you have been sh…
· InvestSim / invest-performance-sim (private)
Splitting the Paper Lab: Why InvestSim Lives in Its Own Repo
[Invest AI](https://xingai.app/apps/invest-ai) answers “what does today look like?” — signals, dashboard context, and (eventually) richer decision surfaces. A different question…
· XingAI Invest AI
Shipping Invest AI V1: Runbook Thinking for Fly.io + Vercel
[ADR-004](2026-05-12-v1-hosting-fly-io.md) explains why we picked Fly.io for the Python side and Vercel for the Next.js side. It does not replace copy-pasteable commands, projec…
· XingAI Invest AI
Why We Flipped V1 Routing to OpenAI-First (and What Comes Next)
Early V1 tried providers in Ollama → Gemini → OpenAI order. The idea: avoid cloud cost until you must.
· XingAI Invest AI
Treating LLM Output Like Cache Rows (Planned): Worker-Owned Inference
This post documents a decision we accepted, not something already running in production. It belongs in the same story as our CQRS market cache: derived data should have one writer.
· XingAI Invest AI
Five Layers of “Not Investment Advice” for an AI Finance Product
Invest AI surfaces BUY / HOLD / SELL style signals and confidence scores. In many jurisdictions, anything that looks like personalized investment advice can drag you into regula…
· XingAI Invest AI
Capping Free-Tier AI Calls Without Slowing Down Development
Three endpoints burn real tokens on every hit:
· XingAI Invest AI
Four Runtimes, One Repo: Why We Skipped Nx (For Now)
Invest AI is not one app — it is four top-level packages:
· XingAI Invest AI
CQRS with SQLite: One Writer, Many Readers, No Drama
Market dashboards want fresh data. Upstream APIs want low call volume. Users want sub-200ms first paint.
· XingAI Invest AI
Hosting an AI Side Project for $0: How We Picked Fly.io for Invest AI V1
We had the V1 of Invest AI working locally. The frontend was already on Vercel — that part is solved for everyone. The harder question:
· XingAI Invest AI
Building a Three-Layer AI Architecture for Investment Decisions
Most AI investment tools today work like this: dump market data into one LLM, get an answer back. It's simple, and it's wrong.
· Invest AI · `stock-ai-monitor`
The Monitor That Wouldn't Stop Refreshing — A React Effect Loop Postmortem
Project: Invest AI · stock-ai-monitor
· XingAI Invest AI
MCP Phased Rollout: From Dashboard to Autonomous Trading
When you design an AI investment platform, the dream architecture looks like this:
· XingAI Invest AI
Why We Chose a Hybrid LLM Pipeline: Gemini Screens, OpenAI Decides
Our V1 system uses a fallback chain: try Ollama (local, free), then Gemini, then OpenAI. Whichever model responds first wins. It's resilient, but every model does the exact same…