We build AI systems that
hold up in production

ArqiSoft is a software engineering company. We build agent systems and the data platforms underneath them for federal, research and commercial teams, and we've been shipping data and software platforms since 2018.

Let's
talk
See our work →
7 days → 5 min
deployment time on a federal data collection system we re-architected
3 days → 5 hrs
release cycle on Leanda, our open science data platform
2 months
from idea to a working ML proof of concept for a fintech product

What we build

Agent systems

Multi-agent systems wired into your tools and data, with explicit state and routing, and human checkpoints where they matter.

Evals and guardrails

Failure analysis, eval suites that run in CI, guardrails and audit trails, so you find out an agent is wrong before your users do.

Data platforms for AI

Lakehouse, streaming and retrieval pipelines that agents can rely on, with data contracts, lineage and access controls.

Modernization

Legacy monoliths moved to cloud-native services, with the tests and deployment pipeline to keep them moving.

Selected work

Modernizing a federal data collection system

U.S. Department of Education
A 15-year-old monolith slowed every release. ArqiSoft led a team of 16 engineers re-architecting it into cloud microservices over 18 months.
  • Deployments from 7 days to 5 minutes
  • Feature delivery from weeks to under 5 days
  • Documented REST APIs and integration testing
  • Production support after cutover

Leanda.io: open science data platform

Open source, built with NIH subject-matter experts
Research teams needed one place to ingest, process and model data in many scientific formats. We built an open-source platform with ML training and prediction pipelines.
  • 10+ scientific data formats
  • Release cycle from 3 days to 5 hours
  • Microservices, Docker or AWS CloudFormation deployment
Leanda.io on LinkedIn Leanda.io on GitHub

EZ Assets: AI-assisted investing

Fintech product build
A platform that ranks investment assets with gradient-boosted models trained on historical market data and serves predictions to the product in real time.
  • Proof of concept in 2 months
  • Beta with hundreds of users in 6 months
  • Team of 7: ML engineers and developers

Open source

We publish what we can. The code is the easiest way to see how we work.

Data platform · since 2018

Leanda

An open science data platform: ingest, process and model data across 10+ scientific formats, with ML training and prediction pipelines. Built with NIH subject-matter experts and still actively developed.

View on GitHub →
Agent systems · reference implementation

ai-team

One nine-agent software team that runs on LangGraph, CrewAI or the Claude Agent SDK behind one interface, so every failure can be traced to the model, the framework or the harness. Failure taxonomy, an eval gate in CI, per-role cost tracking, about 1,900 tests.

View on GitHub →
Data platforms · benchmark

Iceberg performance benchmark

A test suite that runs all 99 TPC-DS queries against four Snowflake table formats: native, Iceberg with the Snowflake catalog, Iceberg with AWS Glue, and external Parquet tables. It measures time, data scanned, cache hits and cost per query, with Terraform for the infrastructure.

View on GitHub →
Training · MIT licensed

Learn Python

A free, hands-on Python course that explains why, not just what: notebook lessons, how Python works inside, algorithm patterns, timed challenges and small projects, with an optional AI tutor mode.

View on GitHub →

About ArqiSoft

ArqiSoft is a software engineering company in Rockville, MD, founded in 2018. We started with data platforms for federal agencies and research organizations, and today most of our work is AI agent systems and the data they depend on. Clients have included RTI International, Vertivis, Technopeak and Arome Science.

How we work

Start from the failures

We run the system and read the traces before proposing anything. Most fixes come from knowing exactly where it breaks.

Measure, then change

Every change ships with the check that would have caught the problem: an eval, a test, a gate in CI.

Built to be audited

Federal and research work taught us to log every action, keep permissions tight and make decisions traceable.

Your code, your cloud

What we build lives in your repositories and your accounts, documented so your team can own it after we leave.

Capabilities

AI Engineering

  • Agent systems and multi-agent orchestration
  • LLM integration, context engineering, memory and RAG pipelines
  • Evals, failure analysis and regression gates in CI
  • Guardrails, guarded writes, audit trails and spend limits
  • Data engineering and real-time data pipelines
  • Microservice and cloud-native architecture
  • Vector and general-purpose data stores
  • Tracing, monitoring and cost tracking for AI systems

AI Training

  • For engineers: a multi-session course on building agent systems, from orchestration and context to evals and guardrails
  • For non-technical teams: a 4-week program with a participant workbook
  • Concepts first; hands-on labs are optional and kept separate
  • Tailored to your tools and your own use cases

Tools and Technologies

LLMs
Anthropic Claude OpenAI Hugging Face Llama Mistral
Agentic
MCP LangGraph Agent SDK CrewAI
RAG & Vector
Pinecone pgvector Weaviate Chroma
Data Engineering
Spark Airflow dbt Kafka Snowflake Apache Iceberg
Cloud & Infra
AWS GCP Docker Kubernetes Terraform
Languages & DB
Python TypeScript PostgreSQL MongoDB Redis

Testimonials

Bongjun

ArqiSoft team helped us to migrate our legacy app to modern architecture. We were really happy with the way their team performed and communicated. We would definitely contract them again for the future projects!

Nabil

We had a multi-year contract with ArqiSoft and were satisfied with how easy it was to work with their team. Definitely looking forward to working with this company in the future.

Alexander

We collaborated with ArqiSoft on a product-market fit for a large scale data repository. Working with these guys was always easy.

Clients and collaborators

Tell us what you're building

Book a 30-minute intro call, or send us a short note. We reply within one business day.

Book a call info@arqisoft.com