Co-Founder & CTO · Propstreet AB
Venture-backed AI-native B2B platform for commercial real estate · Stockholm / Brussels
Operational CTO through Aug 2026 · board member from Sep 2026
- Principal interface between engineering and business users: mapped broker, investor and domain-team workflows, turned recurring field needs into prioritised AI use cases, and guided adoption from first release to daily production use.
- Designed and shipped Propstreet Agent on the OpenAI Responses API: streaming multi-step execution, parallel tool calls, 70+ authorization-gated tools, strict structured outputs and single-flight concurrency.
- Implemented safeguards against prompt injection, data leakage and hallucination propagation through permission-aware tool access, approval gates, schema-enforced outputs and operational guardrails.
- Built the retrieval and research stack: embeddings, vector search, reranking and calibrated scoring; hosted web search; progressive-disclosure skills; vision/OCR over deal documents.
- Made the platform measurable with per-request cost ledgers, quotas, budgets, retries, kill switches, prompt-cache accounting and OpenTelemetry GenAI traces.
- Established eval-driven development: 41 promptfoo suites and ~1,000 deterministic, JSON-schema and LLM-judged assertions tied to production prompt snapshots, gating prompt and model rollouts.
- Codified reusable patterns into an external MCP server, 21 versioned public APIs, four MIT-licensed tools and scheduled or event-triggered agents.
- Built and operated the full B2B SaaS across C#/.NET, TypeScript/Vue and Azure, backed by 11,000+ automated tests, converting each frontier-model release into customer-facing features.