Emmis Voice AI Assistant
A real-time Swedish voice assistant built for the people digital services keep leaving behind, shipped as a live product with paid plans in place, on infrastructure she owns end to end.

Portfolio
Production AI agents first, then the transformation work that taught me how systems actually move. Each card opens a deep dive: thesis, design, outcome.
A real-time Swedish voice assistant built for the people digital services keep leaving behind, shipped as a live product with paid plans in place, on infrastructure she owns end to end.

A two-tier customer-care inbox where every message arrives with an AI-drafted reply, grounded in the operation's own knowledge base, gated by code-level guardrails that run before the model ever drafts, and always sent by a human.
A live multi-agent platform that continuously tracks Swedish public sector orgs, their procurements, decision-makers, and LinkedIn activity, and drops ready-to-send outreach drafts straight into the team's inbox every morning.
The AI-first operating layer wrapped around the live Emmis voice assistant. Nine specialised agents handle engineering, finance, marketing, customer care, compliance, security, and quality under a constraint-based governance system. One founder. Continuous, automated pipelines.
Each consultant gets their own AI agent, trained on their projects and their expertise, that represents them directly to clients in conversation, instead of a CV.
An early prototype that combined fraud and AML signals into a single pre-investigation brief, the part of an AML officer's day that costs the most time and produces the least insight.
A multi-tenant educational platform that turns a consultancy's proprietary methodology into a live product, interactive workshops, AI analysis on every exercise, RAG-grounded gap analysis against each tenant's own documents, and a maturity self-assessment that follows individuals over time.
An AI agent representing Alfija, the front door to her work, instead of a CV.
A continuously-running personal health intelligence agent she designed and runs against her own body, twelve health domains observed, multimodal natural-language input over Telegram, bodyweight training designed against its own session history, domain science codified at the schema level.
A persona-engineered personal AI companion she designed and runs on her own server, voice in, voice out, persistent memory across days, web-grounded, owned end to end.
A fully automated trading-signal system she designed and runs against her own retail portfolio, multiple brokers, multiple timeframes, one Monday-morning routine.