ForwardDeployed
One workspace where Meama's employees and AI agents work on the same company data, with human approval for anything consequential.One workspace for Meama's people and AI agents.
Full national scholarship
I work inside operations teams, find the work that still runs on spreadsheets, and build the AI systems that replace it. My research asks why those systems fail and how to make them dependable.I build the AI systems operations teams run on. My research is about making them dependable.
CADNS
Meama · Company operating systemA shared workspace where Meama's employees and AI agents work on the same company data. Underneath is one data foundation: a warehouse, a metrics layer and access control. On top, people ask for work in a chat, and agents act with the permissions of the person who asked. Consequential actions wait for human approval. I mapped workflows across 15 departments to decide what to build first. In development.One workspace where employees and AI agents share company data. Agents act with the asker's permissions; big actions need human approval. In development.
Statica
Meama · Call-centre intelligenceTranscribes and classifies calls to Meama's call centre, pulls the exact customer quotes behind each classification, and tracks every commitment made to a customer by owner, deadline and completion. Unclear cases go to a person, and their decisions improve the next classification.Classifies call-centre calls, quotes the evidence, and tracks every promise to a customer until it's kept.
Python · LLM APIs · Speech-to-text · PostgreSQLWilco
Lead Software Engineer · EV taxi platformI lead a team of three building Wilco, an all-electric taxi platform for Tbilisi. It is three products on one backend: a rider app, a driver app and an admin panel, plus charging software for the fleet's own stations, where drivers plug in and are billed per kWh. In development.Leading a team of three on an all-electric taxi platform for Tbilisi: rider app, driver app, admin panel and fleet charging.
PRMTR
Co-founder & CTO · Commercial analyticsAI agents investigate sales and marketing data and propose actions; a person approves each one. I designed the reliability layer: an independent verifier re-runs the SQL behind every finding, and pricing and margin rules stay in ordinary code, not in the model. Live at four enterprise clients, with 220K+ orders processed and $50K in attributed revenue in the first two weeks.AI agents propose commercial actions and a person approves each. A verifier re-runs the SQL behind every finding. Live at four enterprise clients.
Python · FastAPI · PostgreSQL · LLM agentsGrid Park
ARCI Engineering Hub · Operations platformI designed the database for an operations platform covering orders, dispatch, fleet, billing and pricing. Business rules are stored as data, so pricing and workflows can change without new code.Database for an operations platform: orders, dispatch, fleet, billing, pricing. Business rules live as data.
PostgreSQL · Supabase · SQL