Pantheon Research

The Pantheon Research Master Coverage
Pantheon Research High Level Architecture
Overview Cross-asset research command center
Macro Framework Overview
Macro Framework Reconstructed signal backtest
Universal Technical Analysis Weekly Scanner
NVDA Price · Fair Value · Signal Timeline
BTC Price & Signal History
Technical Analysis Signal Chart
Alibaba Ticker Profile
Guru Council Matrix
Macro Signal Weekly Telegram report
FICC Signal Weekly Telegram report
Stock Technical Analysis Signal Telegram report
ETH Valuation Weekly Telegram report

Inspiration

Pantheon Research was inspired by a simple but painful problem: modern investors have access to more market data than ever, but the actual research workflow is still fragmented.

To understand one asset, an investor may need to check macro indicators, price charts, valuation metrics, earnings data, news, sentiment, crypto liquidity, backtest results, and personal notes across many different tools. The hard part is not only collecting data, but turning fragmented information into a coherent investment view.

Pantheon Research started from one question:

What would an AI-native investment research operating system look like if it combined structured financial data, institutional-style investment frameworks, deterministic signal engines, and LLM-powered reasoning?

The goal is not to build another finance chatbot or an AI stock picker. The goal is to build a disciplined research platform where AI helps investors reason from evidence, understand uncertainty, and make better human decisions.

The deeper inspiration is that AI can give an individual investor something that used to require a full institutional research team: systematic frameworks, continuous monitoring, data pipelines, research summaries, signal distribution, model comparison, and eventually execution support.

Pantheon Research is built around one core belief:

AI should not replace the investor. AI should compound the investor's discipline.

In simple terms, the thesis is:

$$ \mathrm{Wrong\ Strategy} \times \mathrm{AI} = \mathrm{Faster\ Loss} $$

$$ \mathrm{Right\ Strategy} \times \mathrm{AI} = \mathrm{Compounded\ Discipline} $$

What it does

Pantheon Research is an AI-native investment research platform for public markets. It helps users analyze macro conditions, equities, crypto assets, DeFi yield, fixed income, foreign exchange, commodities, technical signals, and prediction markets in one unified system.

The platform combines:

Structured market data
Macro and cross-asset indicators
Equity fundamentals and technical signals
BTC and ETH cycle research
DeFi stablecoin yield monitoring
Fixed income, FX, and commodity dashboards
Technical analysis / Market Pulse scanning
Prediction market research
Backtest analytics
Data quality and provider health checks
Telegram / Slack signal delivery
Qwen and DeepSeek qualitative research overlays

Pantheon Research is organized as a four-layer investment AI stack:

Strategy Layer Institutional-style frameworks across macro, equities, crypto, DeFi, FICC, technical analysis, and prediction markets.
Information Layer Dashboards, data plumbing, canonical databases, APIs, data quality checks, and research artifacts.
Signal Layer Telegram bots, Slack workflows, alerts, and AI-generated briefs that distribute scored research signals.
Trading Layer A roadmap from manual decision-making to paper trading, broker integration, human-approved execution, and eventually constraint-bound automation.

The workflow is:

$$ \mathrm{Raw\ Market\ Data} \rightarrow \mathrm{Canonical\ Snapshot} \rightarrow \mathrm{Deterministic\ Signal} \rightarrow \mathrm{LLM\ Overlay} \rightarrow \mathrm{Human\ Decision} $$

For the Qwen Cloud Hackathon, the most important new feature is the Qwen + DeepSeek dual-model qualitative overlay. Qwen Max and DeepSeek analyze the same structured equity evidence packs independently, using the same schema, and their outputs are compared factor by factor.

The system does not ask an LLM to directly decide what to buy. Instead, it first builds a structured evidence layer, then uses Qwen and DeepSeek to analyze business quality, moat, pricing power, capital allocation, red flags, and missing evidence.

How we built it

Pantheon Research was built as a full-stack research platform with a modular architecture.

The frontend is a responsive web dashboard built with React, TypeScript, and Vite. It organizes different research domains into clear modules, including Global Macro, Equities, Crypto, DeFi Yield, FICC, Technical Analysis, Prediction Markets, and Research Ops.

The backend is built with Python / FastAPI-style services. It provides API endpoints for market data, signal generation, backtest analytics, research snapshots, Qwen / LLM outputs, and admin operations.

The data layer follows a DB-first architecture. Market observations, provider snapshots, derived signals, and research artifacts are stored in PostgreSQL-style databases before being used by the application.

The core architecture looks like this:

Market Data / APIs / Scrapers / Filings
        ↓
Data Ingestion + Normalization
        ↓
Canonical Research Database
        ↓
Derived Snapshots + Evidence Packs
        ↓
Deterministic Signal Engines
        ↓
Qwen / DeepSeek Qualitative Overlay
        ↓
Dashboard + Telegram + Slack + API
        ↓
Human Investment Decision Workflow

The current stack includes:

React / TypeScript / Vite for the frontend dashboard
Python / FastAPI for backend services and APIs
PostgreSQL / Railway PostgreSQL / Alibaba RDS for structured storage
Vercel for the primary frontend deployment
Railway for the primary backend / API deployment
Alibaba Cloud for the Qwen hackathon deployment path
Nginx + Dockerized FastAPI backend on the Alibaba Cloud deployment
Qwen Max / DashScope for qualitative overlay and model comparison
DeepSeek as a parallel qualitative research model
Telegram and Slack workflows for signal delivery and research interaction

The Alibaba Cloud deployment is now working as a demo-ready path. The frontend, backend, database connection, and major dashboard flows were repaired and verified for the hackathon demo. The stack now supports a live Alibaba Cloud deployment with Qwen integration:

Alibaba Cloud
    ↓
Nginx Frontend / Reverse Proxy
    ↓
Dockerized FastAPI Backend
    ↓
Alibaba RDS
    ↓
Qwen / DashScope Qualitative Overlay

A key design principle is the separation between deterministic signals and AI interpretation.

The deterministic engine owns:

Scores
Gates
Regimes
Hard stops
Data quality labels
Validation states
Framework verdicts

The LLM overlay owns:

Business quality analysis
Moat interpretation
Pricing power review
Capital allocation review
Risk and red-flag synthesis
Missing evidence detection

This separation reduces hallucination risk. The LLM is not the source of truth. The data and framework are the source of truth. Qwen and DeepSeek act as analyst layers on top of structured evidence.

Qwen integration

For the Qwen Cloud Hackathon, Pantheon Research integrated Qwen through Alibaba Cloud Model Studio / DashScope as a core qualitative research overlay model.

The Qwen workflow works as follows:

Ticker Evidence Pack
        ↓
Qwen Analysis
        ↓
Schema Validation
        ↓
Stored Research Snapshot
        ↓
Comparison with DeepSeek Output
        ↓
Agreement / Divergence / Evidence Gap

Qwen and DeepSeek are treated as independent research analysts. They receive the same structured evidence pack, use the same output schema, and are stored separately. The comparison engine then evaluates where the models agree, where they diverge, and where the evidence is not strong enough.

Current Qwen overlay results include:

Metric	Result
Qwen comparison-capable coverage	312 tickers
Markets covered	US / China / Hong Kong / Singapore
Healthy model comparisons	312 / 312
Qwen market split	US 117 / CN 69 / HK 103 / SG 23
DeepSeek baseline universe	1,331 tickers
Persist recovery result	16 / 16 recovered
Full-universe parity status	Not pursued; low-liquidity tail intentionally excluded

The purpose is not to decide which model is “right.” The purpose is to detect overconfidence, missing evidence, and model disagreement before a human investor makes a decision.

This is especially valuable in finance, where a confident unsupported answer can be more dangerous than no answer.

Challenges we ran into

The biggest challenge was data reliability.

Financial data is fragmented, inconsistent, and often incomplete. Different providers have different coverage, refresh frequency, symbol formats, rate limits, and data quality. This made it necessary to build data gap audits, provider health checks, fallback routes, canonical database tables, and explicit freshness labels.

During the Alibaba Cloud deployment work, we faced real production-style issues:

Environment mismatch between Railway and Alibaba deployments
Macro data returning mock-mode payloads despite valid DB snapshots
Market Pulse / Technical Analysis becoming stale because provider authentication and Yahoo access behaved differently on Alibaba Cloud
Provider rate limits causing equity rows to be skipped
BTC data becoming partially degraded because some price and macro inputs were not configured in the Alibaba environment
Frontend route and data-state issues that had to be repaired for a reliable demo path

These challenges became part of the product design. Pantheon Research does not hide degraded states. It labels them.

For example, the system distinguishes:

Fresh
Stale
Hard stale
Provider degraded
Data gap
Not implemented
Insufficient coverage

Another challenge was preventing AI hallucination. In finance, a confident but unsupported answer is dangerous. Pantheon Research addresses this by grounding AI-generated narratives in structured signals, stored observations, explicit frameworks, and schema-validated evidence packs.

A third challenge was product complexity. The platform covers macro, equities, crypto, DeFi, fixed income, FX, commodities, technical analysis, and prediction markets. Without a clear architecture, the product could easily become overwhelming. We solved this by organizing the system into the four-layer stack:

Strategy → Information → Signal → Trading

We also had to think carefully about execution. Signals are not the same as trades. A signal may tell us direction, risk level, or position cap, but trading requires order routing, slippage control, reconciliation, attribution, broker integration, and risk overrides.

That is why Pantheon Research intentionally builds trading last, after the research framework, data layer, and signal layer are stable.

Accomplishments that we're proud of

We are proud that Pantheon Research became more than a demo. It is a real research operating system with multiple asset classes, production APIs, data pipelines, admin controls, signal bots, and AI-powered interpretation.

Some accomplishments we are especially proud of:

Built a unified research dashboard across macro, equities, crypto, DeFi, FICC, technical analysis, and prediction markets.
Created a four-layer investment AI architecture: Strategy, Information, Signal, and Trading.
Built institutional-style frameworks for macro, US equities, China equities, Singapore equities, BTC, ETH, DeFi yield, fixed income, FX, commodities, technical analysis, and prediction markets.
Created a DB-first architecture for more reliable market research.
Integrated deterministic signal engines with Qwen and DeepSeek qualitative overlays.
Built a dual-model comparison workflow across US, China, Hong Kong, and Singapore equities.
Generated Qwen qualitative overlays for 223 / 223 target equities.
Built an Alibaba Cloud deployment path with Nginx, Dockerized FastAPI, Alibaba RDS, and Qwen / DashScope integration.
Repaired the Alibaba frontend and made the main demo path usable for the hackathon.
Restored Macro and Market Pulse / Technical Analysis readiness on the Alibaba deployment.
Built Telegram and Slack workflows for signal delivery and research briefs.
Built Research Ops tools for data quality, provider health, readiness monitoring, and data gap audits.
Added backtest analytics to evaluate whether signals have historical usefulness.
Designed the trading roadmap carefully, starting with manual decisions and paper trading before any automated execution.
Supported bilingual research workflows for English and Chinese-speaking users.
Designed the platform around explainability rather than black-box AI output.

Most importantly, Pantheon Research demonstrates that AI in finance should not just generate text. It should sit on top of a structured research system and help users reason from evidence.

What we learned

We learned that useful financial AI is mainly an infrastructure problem.

A strong model is important, but the quality of the output depends heavily on the quality of the data, the structure of the signals, and the clarity of the research framework.

We also learned that deterministic systems and AI systems are complementary.

Deterministic pipelines are better for:

Calculation
Normalization
Scoring
Backtesting
Consistency
Auditability
Fail-closed behavior

AI systems are better for:

Summarization
Explanation
Translation
Contradiction analysis
Business-quality reasoning
Evidence synthesis
Model-to-model comparison

The most important lesson is:

AI should not replace the research process. AI should make the research process more structured, faster, and more explainable.

We also learned that AI trading projects fail when they skip layers. If a project jumps directly from an LLM to trade execution, it becomes a fast loss machine. The correct order is:

Strategy → Information → Signal → Trading

Pantheon Research taught us that the future analyst workflow will likely be a loop between human judgment, data infrastructure, deterministic systems, and AI reasoning.

The human still needs the framework. AI compounds the framework.

What's next for Pantheon Research

The next step for Pantheon Research is to evolve from a research dashboard into a full AI-native investment research and execution stack.

Product-wise, the roadmap follows the four-layer architecture:

Strategy Layer Continue refining institutional-style investment frameworks across macro, equities, crypto, DeFi, FICC, technical analysis, and prediction markets.
Information Layer Improve data coverage, data quality, provider routing, canonical database design, research dashboards, and admin monitoring.
Signal Layer Expand Telegram, Slack, email, API, and AI-agent delivery surfaces so research signals can reach users in real time.
Trading Layer Move carefully from manual decisions to paper trading, broker integration, human-approved execution, and eventually constraint-bound automation.

Near-term product priorities include:

Expanding Qwen-powered research overlays beyond the current 223-equity target universe.
Adding sector-specific and multilingual CN/HK prompts.
Improving Qwen vs DeepSeek model-comparison diagnostics.
Strengthening Alibaba Cloud deployment reliability.
Improving data provider routing and freshness monitoring.
Expanding browser-level smoke tests and route-contract checks.
Completing the paper-trade harness where every signal becomes a trackable ticket with P&L attribution.
Improving visualization for valuation, technicals, macro regimes, and historical signal performance.
Building personalized watchlists, alerts, and research briefings.
Strengthening mobile / PWA experience for daily signal consumption.

Commercially, Pantheon Research will explore several business model paths:

Pro research subscription for individual investors and power users.
Premium AI research reports across macro, equities, crypto, DeFi, and FICC.
Signal and alert products delivered through Telegram, Slack, email, or API.
B2B research tools for family offices, crypto funds, independent analysts, and wealth managers.
Data quality and Research Ops tooling for teams building their own AI investment systems.
Enterprise or white-label deployment for financial institutions that need customized AI research infrastructure.

The next business milestone is to move from a builder-led product to a more scalable company structure. This means clarifying the target customer, validating willingness to pay, designing pricing tiers, and identifying which modules are most commercially valuable: macro research, equity research, crypto signals, DeFi yield, or AI-powered cross-asset allocation.

Pantheon Research also plans to build a small founding team around several key roles:

AI / LLM engineer
Full-stack product engineer
Data engineer
Quant / backtesting researcher
Financial research analyst
Growth / business development lead

The long-term vision is for Pantheon Research to become an AI-native investment co-pilot: a system that continuously monitors markets, detects regime changes, explains risks, validates signals, distributes research, and helps investors make more disciplined decisions.

The core belief remains: