01Public procurement · SaaS
Bid documents drafted by agents, approved by people
8 kinds of official source, with new procurement notices read every 30 seconds
Companies bidding for public contracts spend days assembling each tender pack from long, inconsistent documentation, and one missed requirement can disqualify a bid.
Evidence In production with hundreds of users, and it has passed a legal review.
Built It reads the official sources, matches each new tender to the companies it suits and drafts the pack in stages: one agent plans it, writers draft the sections in parallel, and an independent reviewer sends weak work back for repair. The system only drafts; a person reviews and submits every document. In a test on real tenders with 17 planted defects, the reviewer agent caught all 17.
Go · PostgreSQL + pgvector · Kubernetes · Multi-agent pipeline
On a call Ask for a walkthrough of a pack drafted for a sample company.
Architecture, back-end and agent design · client system · in production
02Sales, marketing and legal services
Multilingual voice and chat agents, live with customers
250,000+ minutes of live sessions since May 2026, and more than 40,000 sessions in the 60 days to mid-November
Many customers would rather talk than fill in a form. A voice agent that lags, mishears accents, talks over people or makes up answers is worse than the form, and one left unwatched can run up costs quickly.
Evidence Live since May 2026 in English, Romanian, French and many other widely spoken languages. Monthly use has grown from about 7,000 minutes in May to more than 60,000 in the first half of November.
Built Each agent listens and replies in one continuous stream, waits until the caller has finished speaking, and can call tools mid-conversation. The sales agents hand a conversation to a person when it needs one. Answers draw on the client’s documents and, in some deployments, the web, and each one is marked with its source. The deployments limit how many sessions a visitor can open and how many run at once, and have a daily budget, a kill switch and backup models. The agents are tested against fixed sets of conversations.
LiveKit · Python · ElevenLabs · Retrieval + reranking
On a call Ask for a live demo, and for the controls that keep cost and answers within bounds.
Architecture and build · client system · in production
03Live events and industrial sites · Safety
Camera analytics that never store video
0 video recordings kept, checked by an audit command in the client’s deployment
Event organisers need crowd density and flow in real time. Safety teams need fast alerts when someone is missing protective equipment or goes down. Neither should create a surveillance archive.
Evidence Piloted on site at a large music festival in Romania with the organisers and public authorities, and now going into production for an industrial client. One machine analyses 16 live camera streams.
Built Live streams are analysed in memory on a GPU: crowd counts, flows and alerts for events, and protective-equipment and person-down alerts for industrial sites, shown on a bilingual operator console. No video is recorded. An audit command, run in the client’s deployment, fails if it finds a video file, an unblurred image or an image with no database record behind it. Where evidence is needed, only blurred stills are stored; every access to them is logged, and they are deleted after a fixed period.
Computer vision · TensorRT · Python · TypeScript
On a call Watch me run the no-video audit live, and ask for a demo.
Architecture and build · client system · from on-site pilot to production
04Public records · Romania
Romania’s public records, linked and traced to source
14 million records in one graph, with 36 million links between contracts, companies, people and public bodies
Procurement, company-registry and officials’ declaration records are public, but split across systems and too large to search by hand. An investigation needs every link traced to its source.
Evidence Runs on Romanian procurement, company-registry and declaration data: 6.4 million contracts and 4.1 million companies so far, imported at more than 10,000 rows a second. Every change to the agent must pass a fixed set of 16 cases before it ships.
Built One graph joined from the three sources, every link carrying its source. An agent with 12 tools answers questions by searching the graph, building views on demand and pulling out the records it needs. The agent and the extraction models run on premises, on a GPU server. The agent reaches the data only through a gateway that decides which queries it may run.
Python · PostgreSQL · Neo4j · Self-hosted LLM
On a call Ask for a demo on Romanian procurement data.
Architecture and build · part client work, part own research