Applied AI builder — I build production systems for documents, voice, retrieval, and agentic workflows.
My work sits between product and infrastructure: Python/FastAPI services, LLM/RAG systems, ASR/TTS pipelines, multi-channel agents, and local-first or controlled deployment when privacy, latency, or data residency matter. I focus on messy, high-friction operational data: documents, scans, calls, internal knowledge, and multi-step business workflows.
- Document intelligence for regulated and operations-heavy workflows
- Voice AI, call analytics, and speech interfaces
- Retrieval-grounded systems with structured outputs
- Private / on-prem AI stacks: serving, embeddings, reranking, and observability
- Logistics / regulated paperwork — extraction, retrieval, assisted classification, and schema-bound outputs for document-heavy workflows
- Call-center QA and operations — local-first transcription, scoring, and reporting pipelines
- Long-cycle sales — multilingual, multi-channel agents with grounding, routing, and human handoff
- Private AI infrastructure — deployable inference, embedding/reranking services, and production-minded system design
| Repository | What it shows |
|---|---|
| VoiceToText | Offline, cross-platform voice-to-text with support for local ASR workflows |
| Scanovich.ai-audio-call | Local-first call analytics: recordings -> transcription -> structured scoring and reporting |
| Services-BGE | Embedding and reranking microservices for retrieval and hybrid search |
| realestate-agent-platform | Multi-channel AI agents with grounding, tenant isolation, and operational hooks |
| ai-agent-tts | Low-latency voice agent stack with streaming speech in/out and dialog control |
| linux-defender | Security-aware Linux operations tooling: audit, scheduling, and monitoring |
More public repos: github.com/FUYOH666?tab=repositories&type=public
A substantial part of my production work lives in private repositories or on-prem deployments. Common themes:
- regulated document workflows
- marketplace analytics and operator tooling
- speech services and bots
- controlled inference and retrieval infrastructure
- internal agent systems and orchestration
- Start from constraints: data shape, latency, privacy, evaluation, and integration surface
- Ship thin vertical slices first, then harden for production
- Prefer structured outputs, health checks, and observability over demo-only magic
- Use agentic patterns where they help, but optimize for reliability, grounding, and failure handling
- High-signal applied AI / product engineering roles
- Early-stage teams that need a strong builder across product and infrastructure
- Focused partnerships around document AI, voice systems, retrieval, and private deployment
- Email: iamfuyoh@gmail.com
- Telegram: @ScanovichAI
- LinkedIn: aleksandr-mordvinov
- Portfolio: scanovich.ai
- If you're reaching out, include the role or use case, data shape, latency requirements, privacy constraints, and deployment expectations.