03 · Broker-neutral investment intelligence
MarketPulse
Designing an evidence-governed research system before committing to product implementation.
How can AI-supported investment research remain attributable, reproducible and under the user’s authority?
- Company context
- Founder, Point Zero AI
- My role
- Requirements, architecture governance, validation and phased implementation direction
- Current status
- Phase 0 complete · Gate 3B pending
- Development approach
- AI-supported implementation under founder-directed requirements, testing and validation.
01 · Problem
The problem at the centre
Investment research is fragmented across sources, tools and time. AI can accelerate synthesis, but without provenance, contradiction handling, history and clear responsibility boundaries it can produce confidence without accountability.
How can AI-supported investment research remain attributable, reproducible and under the user’s authority?
Everyday investors who want more disciplined research and decision review while retaining every final investment decision.
MarketPulse is being defined as a broker-neutral AI Investment Intelligence Agent. Its controlled Founder MVP is an evidence-led research workspace built around attributable sources, living theses, risk visibility, validation, audit history and reproducible reports.
No trade execution, order submission, brokerage credentials or money movement.
02 · Product Direction
A product boundary before a feature list
MarketPulse is being defined as a broker-neutral AI Investment Intelligence Agent. Its controlled Founder MVP is an evidence-led research workspace built around attributable sources, living theses, risk visibility, validation, audit history and reproducible reports.
Product direction
Broker-neutral research and decision support with a permanent user-authority boundary.
Founder MVP architecture
Browser experience, FastAPI modular monolith, PostgreSQL system of record and governed content storage.
Provider boundaries
Provider-neutral adapters, durable jobs, least-privilege controls and explicit cost hooks.
AI assurance
Structured outputs, independent validation, evidence links and visible uncertainty.
Phase 0 foundation
Compatibility probes, configuration, architecture rules, deterministic fakes, test runners and local CI.
Product workflows
Research, thesis, report and interface implementation remains outside current authority.
03 · Process and Contribution
How I directed the work
- 01
Defined the product’s responsibility boundary: brokers execute, MarketPulse informs and the user decides.
- 02
Structured the PRD, engineering specification and authorised build specification as linked controlled volumes.
- 03
Approved architecture decisions covering modularity, persistence, providers, durable work, AI validation and reproducibility.
- 04
Created gated implementation and test registers that distinguish approved plans from authorised work.
- 05
Validated the exact Phase 0 toolchain and synthetic foundation before requesting authority to build product workflows.
I used AI to support research, documentation, architecture exploration and Phase 0 implementation while keeping requirements, approvals and build authority explicit. No later product phase was treated as implemented simply because it was designed.
Phase 0 is supported by 18 verified build tasks and 11 authorised test records using synthetic, local and provider-disabled evidence. Gate 3B remains a founder decision before named Phase 1 work begins.
MarketPulse shows the clearest change in my process: from feature-led building toward governed requirements, architecture decisions, phased authorisation, test gates, privacy boundaries and disciplined scope control.
04 · Testing and Validation
What supports the current status
Verified from the current local build and test suite.
Phase 0 tasks verified
Compatibility and deterministic foundation only.
authorised test records passed
Synthetic, local and provider-disabled evidence.
build specifications
Approved as a plan; later execution remains unauthorised.
test specifications
Approved as a plan; later execution remains unauthorised.
Phase 0 is complete and the full local gate passes. The result is a verified technical foundation and a Conditional GO recommendation for founder review—not a claim that the Founder MVP product is operating.
05 · System
How the system supports the experience
Simplified from current source structure and architecture records. Sensitive implementation detail is intentionally excluded.
06 · Key Decisions
Boundaries that shaped the product
- No trade execution, order submission, brokerage credentials or money movement.
- No live providers, real sensitive founder data or non-loopback access in Phase 0.
- The wider PRD remains under founder review; approved architecture does not imply an operating platform.
- Gate 3B still requires a founder decision before any named Phase 1 work can begin.
07 · Lessons Learned
What I will carry forward
Architecture approval and implementation are different states. Good product communication makes that distinction visible.
For consequential AI, evidence lineage, contradiction handling and reproducibility belong in the product model—not in a disclaimer.
Scope discipline can be a form of velocity: validating the riskiest assumptions early avoids building on an unproven stack.
Technology
Tools used in the build
Technologies used in the current implementation.