MyAlphaEdge

MyAlphaEdge is a market-intelligence platform that aggregates and processes macroeconomic, financial, positioning, news and event data to help users understand the directional context of major markets, including gold, the Nasdaq, the DAX and Bitcoin.

It is a decision-support product. It does not place trades, and it is not an autonomous trading system.

I built it as founder and engineer. This case study describes the engineering problem and the architecture at a high level. Scoring formulas, data-provider credentials, model prompts and internal business logic are proprietary and are not described.

Market context is spread across sources that are structurally incompatible: economic releases on fixed calendars, positioning reports on weekly cycles, price data continuously, and news continuously and in unbounded volume.

Reading all of it is not a realistic daily activity. The practical problem is not access to data — most of it is public — but reconciling sources that arrive on different schedules, in different formats, with different reliability.

• Heterogeneous sources with incompatible update frequencies
• News volume far exceeds what can be read directly
• Signals are individually weak and only meaningful in combination
• Presentation must not imply a certainty the underlying data does not support

• Ingest heterogeneous financial, macroeconomic, positioning and news data on a reliable schedule
• Normalise sources into a comparable internal representation
• Use AI-assisted filtering, classification and summarisation to make news volume tractable
• Produce interpretable scores with visible supporting signals rather than an opaque output
• Present directional context honestly, including when signals conflict
• Operate the whole system as a small, maintainable product

Ingestion is decoupled from serving. Scheduled jobs write normalised records; the API reads from that store. A provider outage therefore degrades freshness rather than availability.

1. 01

   Client Application

   Dashboards and data visualisation

2. 02

   Authenticated API Layer

   Authentication, access control, request validation

3. 03

   Business and Workflow Services

   Score computation and signal aggregation

4. 04

   Database and External Integrations

   Normalised market, macro and news records

5. 05

   Monitoring, Notifications and Background Jobs

   Scheduled ingestion and processing

• Precomputation on a schedule rather than on request, so the interface stays fast
• AI processing applied only to material that passes a cheaper filter first, bounding cost
• Caching of stable derived values
• Batch processing where provider limits make per-item requests wasteful

• Managed cloud hosting with separated environments
• Scheduled jobs run independently of the web tier
• Secrets and provider credentials injected through the environment
• Automated deployment from version control

• Every ingestion run is idempotent and safe to replay
• Per-source failure isolation: one unavailable provider does not stop the pipeline
• Stale-data detection so the interface can state when a source has not updated
• Retry with backoff against provider rate limits
• Authentication and per-user access control on all data endpoints
• Monitoring on job completion, not only on job start

Evaluating a market-context product responsibly is difficult, and overstating evaluation would be worse than stating none. The methodology being developed focuses on data quality and pipeline reliability — completeness, freshness, and correct handling of revisions — rather than on any claim of predictive performance.

No performance, accuracy or return figures are published. The platform is not presented as a predictive or trading system, and no such claims are made.

Benchmarking in progress. Results will be published with the hardware, model, runtime and measurement conditions that produced them.

• Ingestion reliability dominates everything downstream. A sophisticated aggregation over incomplete data is worse than a simple one over complete data.
• Handling source revisions correctly has to be designed in from the first version; retrofitting it means reprocessing history.
• Careful product language is an engineering concern in this domain, not a marketing one — the interface itself must not imply certainty.
• Bounding AI processing cost with a cheap pre-filter mattered more than choosing the strongest model.

• Broader source coverage
• Stronger data-quality reporting surfaced to the user
• Continued work on interpretability of aggregated signals
• Reliability and monitoring improvements as coverage grows

Described at a general level; internal implementation is proprietary.