A search-first global intelligence platform for understanding what is happening in the world, why it matters, and what it could impact next.
Sphere is a full-stack investigation platform that combines a photorealistic 3D globe, live global signals, market intelligence, calibrated retrieval, causal reasoning, and portfolio-impact analysis.
It is designed around one simple interaction:
Ask a question → resolve the entity → retrieve scoped evidence → explain the cause → show market/portfolio impact → save/share the investigation
Example questions:
- Why is TSLA down?
- Compare oil yesterday vs today
- Compare Japan vs Korea
- What does a Red Sea disruption mean for energy markets?
- What changed in Japan in the last 24 hours?
Add screenshots or GIFs here before publishing.
| View | Placeholder |
|---|---|
| 🌍 Hero Globe |  /> |
| 🔎 Search Investigation |  |
| 🔗 Causal Chain |  |
| 📈 Market Tape + Charts |  |
| 🧭 Compare Mode |  |
Sphere turns open-ended questions into structured investigations.
Instead of showing weather, markets, news, flights, health, and conflict as isolated widgets, Sphere connects them into one intelligence workflow.
query
→ intent classification
→ entity resolution
→ scoped retrieval
→ calibrated ranking
→ causal chain generation
→ market posture
→ portfolio impact
→ evidence + caveats
→ save / share / alert
A user asks:
Why is oil up?
Sphere can resolve the query as a commodity question, retrieve oil-related evidence, rank the top drivers, build causal chains, and explain the possible transmission path.
Shipping disruption → supply-chain pressure → oil supply risk → commodity posture changes
The goal is not just to answer what happened.
The goal is to explain:
- what changed
- why it matters
- what it affects
- how confident the system is
- what evidence supports it
- Understand market moves beyond price charts
- Connect global news to financial risk
- Track countries, commodities, FX, and equities in one place
- Save and share structured investigations
- Monitor geopolitical, weather, health, and logistics risk
- Track exposure to countries, sectors, commodities, and currencies
- Build operational dashboards for supply-chain and market risk
- Convert scattered signals into decision-ready intelligence
Sphere acts like a lightweight intelligence workstation:
- live global context
- evidence-backed summaries
- causal drivers
- risk posture
- portfolio impact
- alerts
- shareable briefs
Sphere uses a custom Three.js + React Three Fiber globe as the spatial context layer.
The globe is not just decoration. It is designed to visualize live signals and investigation context.
- 🌊 Ocean shader with darker navy tones and controlled specular response
- ☁️ Cloud shell / volumetric cloud layer for atmospheric depth
- 🌅 Atmosphere rim and twilight shading
- 🌃 Night-side city lights and shadow-side ambient visibility
- 🌌 Dense starfield for cinematic space context
- ☀️ Real-time sun direction and lighting driver
- 🧭 Camera focus controller for country/entity search
- 📍 Domain markers for events, conflicts, health signals, markets, and news
✈️ Planned / experimental support for 3D flight arcs and route visualizations
Sphere’s rendering direction is inspired by:
- cinematic Earth visualization
- operational command-center interfaces
- dark glass UI systems
- premium geospatial intelligence tools
The globe is tuned for:
- depth
- atmosphere
- spatial context
- readable overlays
- restrained visual drama
It is not intended to be a scientific Earth simulator. It is a premium intelligence surface.
Sphere follows an operator-first interface.
The UI is built around dense but readable intelligence surfaces:
- 🔎 top command bar
- 🌍 globe stage
- 📡 awareness rail
- 📊 right-side investigation panel
- 📈 market tape
- 🧾 evidence cards
- 🔗 causal chain cards
- 💼 portfolio impact cards
The intended visual language is:
Palantir-style seriousness
+ Bloomberg-style density
+ Apple-level dark glass polish
+ cinematic globe context
- Search first, not menu first
- Evidence before claims
- Confidence and caveats always visible
- Compact signals, not long prose dumps
- No hidden synthetic/live data ambiguity
- Visual hierarchy should guide the analyst from answer → why → evidence
Sphere supports multiple signal domains.
| Domain | What it represents | Example use |
|---|---|---|
| 📈 Markets | Equities, ETFs, market posture | TSLA, NVDA, SPY |
| 💱 FX | Currency pairs and currency pressure | USDJPY, EURUSD |
| 🛢 Commodities | Oil, gold, energy/material signals | crude oil, gold |
| 📰 News | Geolocated global events | GDELT-style event feeds |
| 🌦 Weather | Storms, alerts, operational disruptions | severe weather near logistics regions |
| 🦠 Health | outbreak or health-risk signals | regional health pressure |
| geopolitical and regional risk | conflict markers, risk posture | |
| mobility and route disruption signals | flight arcs, airport disruption markers |
Each signal can carry:
- timestamp
- severity
- domain
- location
- source health
- confidence
- freshness
- related entities
Sphere is built around a hybrid intelligence architecture.
It does not let the LLM invent facts.
Instead, it separates deterministic truth from language generation.
The deterministic layer handles:
- entity resolution
- query intent classification
- time-window parsing
- scoped retrieval
- evidence ranking
- confidence calibration
- causal chain construction
- market posture scoring
- portfolio impact mapping
The LLM layer is optional and bounded.
It is used for:
- rewriting grounded explanations
- summarizing evidence
- improving narrative clarity
- producing analyst-style language
It is not used as the source of truth for:
- prices
- candles
- risk scores
- causal edges
- portfolio impact
- provider health
- confidence values
Sphere’s retrieval flow is designed to avoid vague global fallback answers.
The system performs:
- query classification
- entity resolution
- domain scoping
- time-window parsing
- evidence retrieval
- reranking
- confidence calibration
- caveat generation
Confidence is based on interpretable inputs such as:
- evidence count
- evidence agreement
- recency
- source diversity
- entity-resolution confidence
- feedback signals from query logs
Sphere includes a query logging and calibration system that can capture:
- query text
- intent
- resolved entities
- evidence IDs
- confidence score
- result count
- latency
- user feedback action
This allows future tuning of ranking weights and confidence formulas.
Sphere includes a deterministic causal-chain engine.
It converts evidence into structured paths like:
event / signal → mechanism → affected domain → downstream asset / region / portfolio exposure
Example:
Red Sea disruption
→ shipping route pressure
→ supply-chain risk
→ oil risk premium
→ energy portfolio exposure
- Causal nodes
- Causal edges
- Mechanisms
- Impact direction
- Impact strength
- Confidence
- Caveats
- Source evidence IDs
Every causal chain must be grounded in evidence. Unsupported causality returns caveats instead of fake explanations.
Sphere can map causal chains to a demo or user portfolio.
Example:
Oil supply pressure → energy sector → XOM exposure
The system classifies exposure as:
- Direct — affected symbol matches a holding
- Indirect — affected domain matches sector/asset metadata
- Weak — broad country or macro exposure
The system does not fake P&L or pretend to provide financial advice.
It shows directional exposure and caveats.
Sphere supports market posture analysis using a combination of:
- technical indicators
- semantic/news pressure
- macro/entity relevance
- confidence and caveats
The posture output can include:
- Strong Sell
- Sell
- Neutral
- Buy
- Strong Buy
Sphere clearly distinguishes between:
LIVECACHEDSYNTHETIC DEMOUNAVAILABLERATE LIMITED
If no live provider key is configured, charts run with deterministic synthetic demo data and are labeled accordingly.
┌─────────────────────────────────────┐
│ Frontend │
│ Next.js + React + TypeScript │
│ React Three Fiber / Three.js Globe │
│ Zustand State Stores │
│ Lightweight Charts │
└──────────────────┬──────────────────┘
│
▼
┌─────────────────────────────────────┐
│ Backend │
│ FastAPI + Pydantic │
│ Intelligence Runtime │
│ Retrieval Orchestrator │
│ Calibration + Reranker │
│ Causal Chain Builder │
│ Portfolio Impact Engine │
│ Alert Evaluator │
└──────────────────┬──────────────────┘
│
▼
┌─────────────────────────────────────┐
│ Storage │
│ PostgreSQL + PostGIS │
│ Redis │
│ Alembic Migrations │
└──────────────────┬──────────────────┘
│
▼
┌─────────────────────────────────────┐
│ Optional External Providers │
│ Alpha Vantage │
│ GDELT │
│ Open-Meteo │
│ USGS │
│ Frankfurter │
│ Anthropic Claude │
└─────────────────────────────────────┘
| Layer | Technology |
|---|---|
| Frontend | Next.js, React, TypeScript |
| Globe | Three.js, React Three Fiber, custom shaders |
| State | Zustand |
| Charts | Lightweight charting surface |
| Backend | FastAPI, Python, Pydantic |
| Persistence | PostgreSQL, PostGIS, Alembic |
| Cache / Alerts | Redis |
| AI Narrative | Anthropic Claude, optional |
| Market Data | Alpha Vantage, Polygon |
| Testing | Pytest, Vitest, Playwright |
| Infra | Docker Compose, health checks |
| Deployment | Railway relay, Cloudfare https://thesphere.icu/ |
# Backend
cd backend
uv run pytest
# Frontend
cd frontend
pnpm test
# Frontend typecheck
cd frontend
pnpm tsc --noEmit
# Optional Playwright
cd frontend
pnpm playwright testThe project includes tests for:
- retrieval
- entity resolution
- compare mode
- time-window parsing
- causal chain generation
- market posture
- portfolio impact
- alerts
- saved investigations
- frontend panels
- globe interaction paths
cp .env.example .env
# Add optional API keys if available
docker compose up --buildFrontend:
http://localhost:3000
Backend:
http://localhost:8000
# Rebuild frontend only
docker compose up -d --build frontend
# Rebuild backend only
docker compose up -d --build backend
# Check Redis
docker compose exec redis redis-cli ping
# Check Postgres tables
docker compose exec postgres psql -U sphere -d sphere -c "\dt"
# Backend logs
docker compose logs backend --tail=200| Variable | Purpose | Required |
|---|---|---|
INTELLIGENCE_DATABASE_URL |
Postgres DSN for investigations/query logs | Recommended |
INTELLIGENCE_REDIS_URL |
Redis URL for alerts/rate limits | Recommended |
INTELLIGENCE_MARKET_DATA_PROVIDER |
Market provider selection | Optional |
INTELLIGENCE_ALPHA_VANTAGE_API_KEY |
Live market candles | Optional |
INTELLIGENCE_ANTHROPIC_API_KEY |
Claude narrative layer | Optional |
NEXT_PUBLIC_API_BASE_URL |
Frontend backend URL | Required |
NEXT_PUBLIC_WS_BASE_URL |
WebSocket base URL if enabled | Optional |
Sphere can run without provider keys, but live functionality improves when keys are configured.
| Phase | Name | What was built | Status |
|---|---|---|---|
| 1 | Foundation | Project setup, initial architecture, frontend/backend skeleton | ✅ Done |
| 2 | Globe Prototype | Early 3D Earth surface and basic scene setup | ✅ Done |
| 3 | Workspace Shell | Initial command-center layout and panels | ✅ Done |
| 4 | Signal Ingestion | Early global signal/event ingestion concepts | ✅ Done |
| 5 | Event Modeling | Normalized event structures and domain concepts | ✅ Done |
| 6 | UI Panels | Early right-panel and rail surfaces | ✅ Done |
| 7 | Globe Visuals | Atmosphere, borders, markers, lighting experiments | ✅ Done |
| 8 | Search Flow | Search-first interaction model | ✅ Done |
| 9 | Intelligence Workspace | Query panel, focus behavior, workspace state | ✅ Done |
| 10 | Globe Shader System | Ocean, atmosphere, clouds, night side, tone mapping | ✅ Done |
| 11 | Live Feeds | Weather, news, markets, health/conflict/feed adapters | ✅ Done |
| 12 | Agentic Investigation | Early agent query service and grounded answer flow | ✅ Done |
| 12.3 | Geographic Trust Repair | Better city/country scoping and fallback honesty | ✅ Done |
| 13 | Portfolio Intelligence | Portfolio surface, holdings context, valuation/posture basics | ✅ Done |
| 13B | Signal Engines | Technical indicators, charting, replay foundation | ✅ Done |
| 14 | Operator UI | Mode system, shell refinement, panel grammar | ✅ Done |
| 15A | Workflow Repair | Onboarding, rail improvements, better demo flow | ✅ Done |
| 15B | Chart Surface | Indicators, technical rating, chart wrapper | ✅ Done |
| 15C | Timeline Intelligence | Trend deltas, what-changed logic, feed warming | ✅ Done |
| 16 | Motion + Market Surface | Ticker tape, replay cursor, hydration fixes | ✅ Done |
| 16.7 | Universal Market Charts | Decoupled charts from portfolio membership | ✅ Done |
| 17A | Posture Engine | Deterministic market posture and provider contracts | ✅ Done |
| 17A.2 | Semantic Market Layer | Semantic/news pressure blended into posture | ✅ Done |
| 17A.3 | Agentic Narrative | Bounded LLM narrative over deterministic posture | ✅ Done |
| 17B | Saved Investigations | Save/restore/share investigation snapshots | ✅ Done |
| 17C | Alerts MVP | Alert rules, alert events, bell surface, rate limits | ✅ Done |
| 18A | Retrieval Orchestrator | EvidenceBundle, time windows, compare planning | ✅ Done |
| 18B | Calibration | Query logs, reranking, confidence calibration | ✅ Done |
| 18C | Scope Enforcement | Entity-first routing, no global fallback, compare fixes | ✅ Done |
| 18D | Causal Chains | Deterministic causal chain engine and top drivers | ✅ Done |
| 19A | Demo Polish | Search examples, story ordering, causal card emphasis | ✅ Done |
| 19B | Portfolio Impact | Causal chain to portfolio/demo-book linkage | ✅ Done |
| 19C | Globe Revamp | Clouds, starfield, real-time sun, marker fixes | ✅ Done |
| 19D | Visual System | Design tokens, gradients, premium dark glass polish | ✅ Done |
| 19E | Launch Verification | Infra audit, provider honesty, chart reliability checks | 🚧 Finalizing |
| 20A | Domain Globe Layers | Flights, health, conflict, news markers/arcs/hotspots | 🚧 In progress / planned |
Most dashboards show data.
Sphere tries to explain meaning.
signal → evidence → cause → impact → confidence → action
This makes it closer to a lightweight intelligence system than a standard analytics dashboard.
Sphere is still a prototype.
Current limitations:
- Not investment advice
- Not a replacement for professional terminals
- Live data depends on provider keys and rate limits
- Some domains use fallback/demo data when providers are unavailable
- No multi-user authentication yet
- No enterprise permission model yet
- Flight/health/conflict visual layers are still evolving
- Globe realism is custom-built and still being tuned
- No CI/CD pipeline yet
- Multi-user workspaces
- Team investigation sharing
- Watchlists and saved entities
- Better onboarding and demo mode
- Advanced alert rules
- Investigation export to PDF / Markdown
- More reliable market providers
- Better flight route providers
- More robust health/conflict feeds
- Historical event replay
- Provider freshness dashboard
- Lightweight RAG over saved investigations
- Deeper causal-chain expansion
- Better confidence calibration from real usage
- Analyst feedback loop
- Domain-specific sub-agents
- Improved domain markers
- 3D flight arcs
- health/conflict hotspots
- better marker occlusion
- improved cloud realism
- more detailed polar views
- CI/CD
- hosted deployment
- auth
- observability
- backups
- rate-limit dashboards
Sphere | Real-Time Geospatial Intelligence Platform
Next.js, TypeScript, React Three Fiber, FastAPI, PostgreSQL/PostGIS, Redis, Docker
Sphere is a real-time geospatial intelligence platform that fuses world signals, market data, calibrated retrieval, causal-chain reasoning, and portfolio-impact analysis into a search-first investigation workflow.
It demonstrates:
- full-stack systems engineering
- frontend visualization engineering
- backend intelligence pipelines
- retrieval and ranking
- causal reasoning
- provider honesty
- production-oriented persistence and caching
Personal project. Shared for portfolio and review purposes. Not licensed for production redistribution.
Built by Anish Nehete.
If you find this project interesting, consider starring the repository ⭐