A selection of work with a portable working artefact:
a webhook contract, a JTBD graph, a metrics framework.
Bevra: agentic content engine behind a webhook seam
AI / SaaS
Goal: unify a solo operator's LinkedIn CRM, discovery, and content generation without coupling release cycles.
Multi-tenant SaaS on FastAPI, React, and Supabase. A self-hosted agentic Next Post Engine
runs on n8n, Claude Haiku 4.5, and pgvector, with human-in-the-loop approval over Telegram.
The two sides talk across three webhook endpoints.
A deterministic runtime eval gate and an offline LangSmith harness both read
criteria.py. Prompt changes get scored against a golden set before they ship.
Working artefacts: the eval-gate penalties, the judge chain, the webhook contract.
Full case study →
Kosovo #MoseMerrMalin: JTBD study for a reforestation campaign
JTBD Discovery
Client: Sustainability Leadership Kosovo. Goal: reframe a B2C reforestation campaign around the jobs different audiences hire it to do.
Five B2C segments identified. Each segment carries a Core Job, a Big Job, and a TAM/SAM/SOM sizing.
Diaspora TAM sized at 600K–950K. Urban-youth TAM sized at 250K+.
Job graph and risky-assumption analysis produced alongside the sizings.
Landing-page copy for the primary segment shipped into the website rewrite at
slkosova.org/mosemerrmalin-welcome.
Working artefacts: the segment × Core-Job × Big-Job matrix, and the copy derived from it.
Giro555: metrics framework for a national charity fundraising alliance
Experimentation
Client: Giro555 (Dutch national charity fundraising alliance). Goal: run campaigns as tests instead of narratives.
Primary, supporting, and counter-metrics defined per test. Counter-metric declared before the test starts.
Results feed a decision dashboard readable by a non-analyst stakeholder.
The framework ports across campaigns without per-campaign rebuilds.
Working artefacts: the metric-triad structure and the up-front counter-metric rule.