Technology
AI & Data
Applied AI + data pipelines: open-weight and frontier models benched by task, evals before demos.
AI & Data — highlighted tech plateOverview
Where it fits.
We ship AI that earns its keep: grounded retrieval, eval-gated releases, and human-in-the-loop fallbacks. Models are chosen by a task bench, not by hype - and customer data stays private by default.
Tools
What we reach for.
Python
3.1xML + data language
PSF Licensedocs →pgvector
0.8embeddings in Postgres
PostgreSQL Licensedocs →LangChain
0.3xLLM orchestration
MITdocs →DuckDB
1.xin-process analytics
MITdocs →ClickHouse
2xanalytics / time-series
Apache-2.0docs →Dagster
1.xpipeline orchestration
Apache-2.0docs →Hugging Face
open-weight models
Apache-2.0 (libs)docs →Frontier APIs
OpenAI/Anthropic; DPA + masking required
commercialdocs →
Use cases
What it's great for.
RAG assistants
grounded answers, citations, eval-gated
Forecasting & anomaly detection
time-series models with human review
Data pipelines
ELT + orchestration with quality gates
Analytics & BI
self-serve metrics, dashboards
Pairings
Plays well with.
- backend developmentcontract-first services + embeddings storage
- cloud devopsmodel-serving infra + GPU pipelines
- databases storagevector + time-series stores
As a service
Where this stack ships.
FAQ
Questions, answered.
Open-weight or frontier models?
Task-based bench: we run open-weight and frontier candidates against your data; the scoreboard decides.
How do you prevent hallucinations?
Grounded RAG with enforced citations, confidence thresholds, and human-in-the-loop fallback - never a hallucination-free guarantee.
Is customer data safe?
Private infra by default. No public API calls without a DPA plus masking; PII is stripped before any external call.
How do you evaluate?
An eval suite (RAGAS-style) gates every change in CI - no deploy on regression.
Do you fine-tune?
Only when RAG + prompting can't meet the bar, and only after it's benched. Fine-tuning is a last resort, not a default.
AI & Data built to fit.
We pick the stack for the job, and justify every choice.