Python
Agent logic, ML pipelines, and most of our document-AI and data work.
Loading…
Proficient across the stack you already run.
Not a logo wall. Every technology here earns its place with one honest line on how we actually use it in agentic work. If it isn't something we've shipped with, it isn't on the page.
Languages
TypeScript
End-to-end across front-ends and typed service layers — strict mode by default.
Go
High-throughput services and tooling where latency and concurrency matter.
SQL
First-class, not an afterthought — schema-fixed query generation drives our analytics agents.
Frameworks & Front-end
Next.js
App Router for SSR/SSG product surfaces and admin consoles.
React / React Native
Web and mobile front-ends, including dual-pane document viewers.
Node.js
Typed API layers and webhook buses alongside the Python services.
FastAPI
Python service endpoints for agent and parsing workloads.
Agent Frameworks
LangGraph
Structured, stateful plan graphs where steps must be observable and resumable.
LangChain
Tooling and retrieval glue — wrapped in the governance layer we build for clients.
CrewAI
Multi-agent role decomposition where it fits the problem shape.
MCP
We expose capabilities as MCP servers so any agent can use them as standard tools.
ML & Serving
PaddleOCR
PP-OCRv5 + PP-StructureV3 for on-prem text recognition and layout analysis.
Table Transformer
Cell-level row/column grids so a comment can snap to the right table cell.
YOLOv10
Fine-tuned to detect engineering markup OCR misses — arrows, callouts, ink.
Qwen2.5-VL
Local vision-language model used as a semantic tiebreaker, fully on-prem.
Whisper-class ASR
Streaming speech recognition for low-latency voice agents.
Data
Postgres
System of record for product apps, with an event log for audit.
ClickHouse
Columnar warehouse for sub-second analytics over billions of events.
FAISS
Vector retrieval over standards libraries and resume archives.
OpenSearch
Full-text and tag-centric traceability alongside vector retrieval.
Airbyte
We build production connectors on the CDK for systems no one else ships — incl. SAP Business One.
Cloud, Infra & Observability
AWS / GCP / Azure
Cloud-native deployments inside your accounts and security boundary.
On-Prem / Air-gapped
A first-class target — full pipelines running with no cloud dependency.
Kubernetes
Orchestration for agent fleets and the services around them.
OpenTelemetry
Structured telemetry from every surface — agents, gates, and plan steps.
HOW WE CHOOSE
Fit-driven, not dogmatic.
We don't lead with a favorite framework. We start from the system: its latency budget, its data gravity, its security boundary, and the team that has to operate it after we leave. The stack follows from that.
Where an open-source tool does the job — PaddleOCR on-prem, ClickHouse for scale, Airbyte for the long-tail connector — we use it and contribute back. Where a managed service removes real operational risk, we use that instead. The goal is a system your team can run, not a monument to a particular technology.
GET STARTED
Does it fit your stack? Let's find out.
Tell us what you run today. A Bytevon engineer will sketch how we'd build on it and meet you where you are — no migration assumed.