TraderLaunchpad
Sole Developer & Architect
Professional trading platform with per-tick market data ingestion, ClickHouse time-series storage, TradingView-powered charting, and AI-assisted trading strategies. Multi-tenant architecture with per-user broker connections, platform-wide price sources, and a hybrid data model where ClickHouse stores finalized M1 candles and Convex handles live in-progress candle state for low-latency UI updates.
M1 canonical bars stored in ClickHouse, aggregated for higher timeframes
Client-side polling with planned WebSocket upgrade
tdrlp-core, pricedata, broker-ledger, widgets, AI, tenant, notifications
Platform + per-user connection model with encrypted tokens
The Problem
Retail traders lack access to the sophisticated tools that institutional traders use. Real-time data, professional charts, and portfolio analytics typically require expensive subscriptions or desktop software. And the data engineering challenge is significant: ingesting per-tick broker data, storing time-series at scale, and serving it to charts in real time.
The Solution
Built a browser-based trading platform with a hybrid data architecture: ClickHouse on a private droplet for historical OHLCV candle storage (MergeTree engine, partitioned by month), and Convex for operational state and live candle updates. TradeLocker broker integration provides market data. TradingView lightweight-charts render the frontend. Multi-tenant isolation via org-scoped broker connections with encrypted credentials.
System Architecture
Technical Decisions
Key architecture decisions and their outcomes
ClickHouse over TimescaleDB for time-series
Needed a columnar store for OHLCV candle queries across thousands of instruments with fast aggregation.
Chose ClickHouse with MergeTree engine, partitioned by month, ordered by (sourceKey, tradableInstrumentId, ts).
Sub-second aggregation queries. SSH-tunneled migrations keep the DB private with zero public exposure.
Hybrid Convex + ClickHouse data model
ClickHouse is great for historical reads but can't provide reactive UI updates for in-progress candles.
Built a priceLiveCandles table in Convex for the current in-progress candle, with finalized candles committed to ClickHouse.
Low-latency chart updates via Convex reactivity, with ClickHouse handling the heavy historical queries.
Convex components for domain isolation
Pricedata, broker connections, ledger, AI, and widgets all have distinct schemas and access patterns.
Used Convex components (mounted in convex.config.ts) so each domain owns its schema, queries, and mutations independently.
Clean domain boundaries. Teams can reason about pricedata without touching broker-ledger code.
Engineering Details
- ClickHouse candles_1m table with MergeTree engine, partitioned by toYYYYMM(ts), ordered by (sourceKey, tradableInstrumentId, ts)
- Hybrid live candle model: Convex priceLiveCandles for in-progress bars, ClickHouse for finalized history
- HTTP-based ClickHouse queries from Convex actions (actions required for external HTTP + timer patterns)
- Broker sync via scheduled polling with lease/metadata tracking (lastSyncAt, syncLeaseUntil, listConnectionsDueForPoll)
- Platform AI settings: daily reports, news alerts, EMA signal/filter roles, strategy AI suggestions
- Multi-tenant: Clerk auth + org-scoped broker connections with encrypted token storage
- TradingView datafeed adapter with custom subscribeBars polling and resolution mapping
Key Highlights
- Real-time TradingView charts with 4-second polling (WebSocket stream planned)
- ClickHouse time-series DB for M1 OHLCV candles on private infrastructure
- Multi-broker integration starting with TradeLocker REST API
- Multi-tenant with per-user and platform-wide broker connections
- AI-assisted trading strategies with OpenAI + custom signal/filter roles
- Convex components for modular domain isolation (pricedata, broker-ledger, widgets)
- Encrypted broker credentials with org-scoped data isolation
- SSH-tunneled ClickHouse migrations for zero-exposure database admin
Tech Stack
Skills & Technologies
Related Articles
Real-Time Everything: Why We Stopped Polling and Never Went Back
Our trading dashboard polled every 5 seconds and users complained about stale data. We rebuilt on Convex with real-time subscriptions and the difference was not incremental — it was a different product.
3D in the Browser: What Actually Works and What Does Not
We embedded a Spline scene in our trading dashboard. The first version ran at 12fps on a MacBook Air. Here is how we shipped 3D that users actually enjoyed instead of endured.