Project · Founding engineer
Rapport
Making conversation intelligence useful for automotive service teams. Read it three ways:
Work
Case studies on the decisions behind the products, then the projects, working demos, research and consulting they came from.
The decisions behind the work, and what changed as a result.
Rebuilding Rapport around manager decisions.A manager dashboard built in seven hours from a written plan, followed by evidence-led investigation, resumable processing, and measured pipeline improvements.Read the case study
45 secThe adoption problem behind the roadmap.Following the work at the service desk to find where adoption breaks, then turning the findings into a ranked roadmap.Read the case study
45 secThe AI feature I stopped building.Why accurate follow-up context deserved attention before automated customer messages.Read the case study
43 secTwo personas. One intelligence layer.The same conversation signals serve private advisor coaching and a manager’s view of the team, with different permissions and next actions.Read the case study
45 secTesting the data before betting on it.Checking historical customer records against the live workflow before committing to an integration strategy.Read the case study
45 secA relationship is more than a transaction.A relationship-scoring prototype using sample data, exploring how conversation signals could help reveal disengagement over time.Read the case study
40 secCustomer intelligence from hours of audio.Separating conversations, speaker roles, promises, and customer risks into evidence people can check and use.Read the case study
40 secWhen the CRM goes down.A transactional outbox, retry backoff, and a circuit breaker keep application writes separate from external CRM delivery.Read the case study
40 secUnblocking a product without the API.An OCR and import path turns the existing appointment sheet into a usable advisor queue while the direct integration remains a future path.Read the case study
40 secOne bad row should not stop the import.Tenant boundaries, duplicate handling, and row-level savepoints isolate failures. The film separates illustrative import data from recorded test results.Read the case study
40 secLetting the intelligence evolve.Backward-compatible extraction schemas are implemented. Versioned raw events and a less coupled ingestion architecture are presented as a proposal.Read the case study
1:45Redesigning the service advisor’s day.The friction in an early tablet workflow, the research behind a redesign, and the interface decisions that followed.Read the case study
1:00The interface became a system.A new customer’s working environment reshaped the product, from desktop capture to a shared visual language.Read the case study
45 secThe data behind the workflow.Looking at documented recommendations and customer decisions to decide what a useful follow-up workflow should support.Read the case study The products, demos, research and consulting behind the case studies.
Project · Founding engineer
Making conversation intelligence useful for automotive service teams. Read it three ways:
Four working demos, each built around one hard part of the same kind of work.
Conversational intelligence
Know why something needs attention: the recorded words behind a flag, and a decision kept on record.
Working demo · synthetic data
Product UX and workflow design
One place to keep the person, the evidence and the next step together, without losing context.
Working demo · synthetic data
Engineering and reliability
Reconnected does not always mean recovered: see what survived, what is still waiting, and what needs a deliberate retry.
Simulator · not deployed
Automotive customer experience
Know what happens next, before you need to ask: clear updates and a choice you can review before confirming.
Working demo · synthetic data
Research
One store’s records showed the gap was in recommendations on record, not in how often customers said yes.
Visual system
Roles, states and contrast decisions for a shared interface language.
Personal piece
The same work told as a drive from San Diego to San Francisco. Optional, and slower.
Consulting
Customer discovery for one team, finding who would buy before anything was built; a faster data pipeline for the other.
Consultant · research and data engineering