Work
Twenty-nine new projects from the last few years — production products, internal systems, brand books, and concepts. Filter by lens.
Showing 29 of 29
SafeTrekr Platform
Trip safety for whole organizations, in one production platform.
Serras
A governed AI delivery system, in closed beta. The abyss is the stage; amber is the decision.
Harley's
A 40-year clothier gets a brand system.
TarvaRI
Public risk feeds turned into trip-specific alerts.
OnDemand Painters
Seven systems consolidated into one field-service product.
TarvaRI Alert Viewer
A map you can zoom from world to alert.
Tarva BXD
Brand decisions you can audit, reasoned against 766 principles.
@tarva/ui
One component library, many brands, light and dark.
Truth Governance Layer
Provenance you can see, before you trust the answer.
SafeTrekr Website
The public case for reviewing every trip.
Discovery Council
Discovery with evidence, gates, and a human at the end.
Tarva Project Room
Where agent work gets evidence, cost, and approval.
Tarva Convergence
Seven repositories, 21 shared packages, one platform.
AgentGen
140+ specialist agents, reproducible and versioned.
Lumos
Not a brand guide — a brand that behaves.
JTN AI Reaper
A credible identity for a new AI product, in days.
Mood Board Creator
Creative direction, tested blind against a control.
EchoQuill
Adapt a manuscript from a source you can defend.
Chronica
Take the tedious out of tabletop.
Skill Resolver MCP
The right skill context, resolved deterministically.
Whitepaper Ensemble Factory
Two weeks of research and publishing, down to hours.
Patent Workflow
Source documents to claim drafts, with consensus checks.
Relevant Compliance
Compliance made legible for people who are not specialists.
Global Christian Relief
Story into action for a global relief nonprofit.
Tarva-ERP for OPē
A manufacturing ERP designed in Figma and taken through a 52-page prototype.
NobleShapes
Repeatable AI image styles, not one-off prompts.
OPē website
Brand assets and a homepage system for a deep-tech company's public face.
SignDesigns.ai
Design a custom sign before you buy it.
ListenBridge.ai
Translation for the moments that are not transactions.
TarvaRI
Public risk feeds turned into trip-specific alerts.
What it is
The intelligence engine behind SafeTrekr: it ingests real-time hazard data from dozens of authoritative public sources, normalizes it, clusters it by place and time, scores it, and routes alerts to the trips they affect.
The problem
A safety analyst cannot watch every feed for every destination every day. Raw feeds disagree on format, geography, and severity, most of it is irrelevant to any given trip, and a late alert is noise.
What we solved
A production pipeline: connectors, custom parsers, a normalized store, geographic and temporal clustering, a risk engine, LLM-assisted triage with human review, and alert routing into SafeTrekr trips. Governance is built in — evaluation harness, replay and shadow mode, worker monitoring, cost logging, drift detection.
My role
I conceived TarvaRI and architected and led its AI-assisted development from conception through deployment; directed IA, UX/UI, brand, design, and development. EIGHT32 added the bridge into SafeTrekr Core.
- Python
- FastAPI
- Supabase
- pgVector
- Redis
- Ollama
- React console
- 40 active source connectors
- 17 categories
- 43 custom parsers
- 5 worker services
40 active source connectors · 17 categories · 43 custom parsers