Founding full-stack engineer
Krystal / KNMI
Founder-stage AI workflow products built with a practical operating mindset: account systems, customer-interaction layers, delivery hardening, and AI-assisted product execution that still required real engineering judgment.
Two product surfaces, handled two different ways.
KNMI is live and can be shown directly in the portfolio. Krystal is also live at `krystaldiscover.com`, but that site explicitly blocks cross-site iframe embedding, so the portfolio presents an internal product surface instead of a broken frame.
Grant discovery, writing support, and portfolio intelligence in one product surface.
Krystal was built to help nonprofits, researchers, and mission-driven teams move from scattered opportunity hunting to a more usable operating flow: discover aligned funding, organize the pipeline, and support the writing process with AI while keeping real product logic underneath it.
Funding pipeline snapshot
Operating lens
Built as an AI-native workflow platform, not a one-shot content generator. The system had to support user trust, subscription state, customer records, and release quality alongside the visible grant-writing layer.
Opportunity queue
Core systems owned
Why this mattered
Krystal fit the broader thesis from the portfolio docs: build systems that expand human capability. In practice, that meant turning grant work into a more navigable, more intelligent workflow instead of leaving users with disconnected search, documents, and manual follow-up.
Customer workflow systems inside an AI product environment.
Krystal and KNMI sit in the space where AI products stop being demo surfaces and start needing real operating logic: account states, subscription controls, data relationships, quality review, and release readiness.
AI products break down when core workflows are under-specified.
The real challenge was not simply calling models. It was building the system around them so users could trust the product, manage their accounts, and rely on the outputs in real workflows.
Architecture, quality control, and founder-level execution.
Core data layer
Built Krystal's database connection layer to centralize customer interactions and support the surrounding account system.
Account management
Implemented practical user workflows such as password changes and subscription opt-out controls.
Engineering review
Reviewed and corrected AI-agent-generated code to keep releases functional, compliant, and maintainable.
Delivery ownership
Handled the CTO layer of the work: architecture choices, release readiness, integration quality, and product tradeoffs.
AI as one layer in a larger product system.
- Customer data and interaction logic organized through a central backend layer.
- Account-state workflows treated as first-class product requirements.
- Model- and agent-driven components reviewed through a production-quality engineering lens.
- Delivery decisions made around reliability and business usefulness, not just feature velocity.
Do not confuse AI velocity with product readiness.
- Build stable data and account foundations before layering on more AI complexity.
- Use AI-generated code where it helps, but keep human engineering review in control of release quality.
- Bias toward workflows users can understand and recover from, especially around account and subscription logic.
Constraints
Founder-stage software must move quickly without collapsing under weak system design, weak account logic, or unreviewed AI output.
Impact
Created a stronger operating foundation for RebrandLand's AI products by tightening the system beneath the visible product layer.
Next
The next step would be deeper observability and clearer product analytics around workflow completion, failure modes, and retention behavior.