Waypoint
Rollouts fail when plans skip policy checks or approvals. Waypoint drafts a sequenced plan, validates it against policy, sends it to the right person for sign off, and backs the plan with shadow run evidence before anything goes live.
- Sequenced rollout plan generated by the agent
- Automated policy check before approval
- Human sign off routing
- Shadow run evidence to build trust before launch
RELAY
Not every request needs the most expensive model. RELAY routes simple requests to a fast, low cost tier and escalates harder ones, then shows the cost, latency and accuracy trade off with a plain language reason for every route.
- Two tier routing between a fast model and an escalated model
- Cost, latency and accuracy deltas for every route
- A 'why this route' explanation for transparency
- Product thinking on unit economics of AI features
BirdSignal
A product concept for location intelligence. It uses a retrieval pipeline over brand configuration and signals, scores its own outputs with an LLM as a judge, and routes each case to automatic action, human review or escalation.
- RAG pipeline over brand configuration and signals
- LLM as a judge quality metrics
- Three way routing: Auto, Human in the loop, Escalate
- Demo figures are illustrative, not real data
Spotcheck
A complete product strategy for connected TV advertisers who need to build creative faster and prove which ads actually drive incremental results. It covers problems, personas, principles, scope, metrics, risks and a roadmap, and is honest about which assumptions are not yet tested with users.
- Problems, personas, principles and scope
- Success metrics, risks and roadmap
- Assumptions clearly labelled as not yet user tested
- Built on experiment based measurement
Assurance Fit
A product concept for recommending the right cover at the right moment. Deterministic eligibility rules decide what is allowed, a weighted relevance score decides what is shown, and the reasoning trace is visible, with open questions listed by function.
- Rules and judgment kept separate by design
- Weighted relevance scoring
- Visible reasoning trace for every recommendation
- Open questions listed by function
Looking for the outcomes from shipped, production products? Those are in the case studies. These projects show how I explore, scope and test AI product ideas before they reach engineering.