Skip to content

About

Sample use cases and workflows for using Briefcase AI

Resources

Code of conduct

Contributing

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Briefcase AI Usage Examples

License: Apache 2.0 Python 3.9+ briefcase-ai v3.2.1

Production examples for the Briefcase AI SDK across financial services, e-commerce, and criminal justice. The repo is organized in two tiers — primitives (domain-neutral, one per file) and compositions (SDK primitives assembled into realistic stories).

Setup

pip install briefcase-ai

Or use the automated setup: ./setup.sh | Docker: ./docker-run.sh build && ./docker-run.sh run

Primitives — the patterns/ library

Single-file, domain-neutral demonstrations of each SDK primitive. Each .py has a sibling .ipynb for notebook users; patterns_walkthrough.ipynb is the unified tour.

# Pattern What it shows
01 Decision capture @capture + DecisionSnapshot — the audit atom
02 Bitemporal evidence BitemporalRecord + InMemoryBitemporalStore
03 Correction append append_correction — corrections never mutate originals
04 Temporal replay AsOfView — clamp reads to a historical instant
05 Versioned policy PolicyRegistry — bitemporal policy as-of
06 Examiner bundle ExaminerBundle — content-addressed, verifiable artifact
07 Cost attribution CostCalculator — per-decision cost + monthly projection
08 Drift detection DriftCalculator — baseline vs. window comparison
09 PII sanitization Sanitizer — detect and redact PII
10 Guardrail pipeline GuardrailPipeline — multi-stage allow/deny
11 Decision replay Regression-test primitive — re-run candidates on stored inputs
12 Rate cards rate_card — price by platform × tier × modifier; batch is half price
13 Prompt-cache cost cache-token billing + CostEstimate.cache_cost; warm-cache breakeven
14 Multi-cloud cost same model across first-party/Bedrock/Vertex/Azure; tier & region levers

See patterns/README.md for the full composition matrix.

Compositions — example suites

Primitives assembled into realistic, industry-specific narratives. The Primitives column lists which pattern numbers each suite composes.

Suite Run Primitives Description
Agentic Payments python agentic-payments/01_bitemporal_basics.py 02+03+04+05+06 7-script walkthrough: cross-border payment routing with bitemporal evidence, versioned policy, examiner bundles
Regulatory Workflows cd regulatory-workflows && python 01_credit_underwriting/example.py 01 everywhere; 02+03+04+06 in the five replay-enabled examples 14 financial compliance workflows (ECOA, BSA, CFPB, SEC, OCC, FINRA). Five include a replay capstone: 01_credit_underwriting, 02_ofac_sanctions, 05_mortgage_fair_lending, 07_aml_transaction_monitoring, 14_algo_trading_surveillance
Vantara Commerce cd vantara-briefcase-demo && python 01_agent_discovery/example.py 01+07+08+12 Agent discovery, cost attribution (SDK pricing + rate cards), drift detection, governance reporting
Criminal Evidence cd criminal-evidence-workflow && python experiments/run_experiment.py --fast 01+09+10+11 LLM evidence summarization with guardrails, replay, PII sanitization, and data lineage

Each suite includes Python scripts, Jupyter notebooks, and detailed documentation.

SDK Quick Reference

from briefcase import DecisionSnapshot, Input, Output, ModelParameters, init
from briefcase.storage import SqliteBackend

init()

decision = DecisionSnapshot("credit_underwriting")
decision.add_input(Input("bureau_score", "685", "integer"))

output = Output("decision", "approve", "string")
output.with_confidence(0.92)
decision.add_output(output)

params = ModelParameters("gpt-4o")
params.with_provider("openai")
decision.with_model_parameters(params)

backend = SqliteBackend.in_memory()
decision_id = backend.save_decision(decision)

# Per-decision cost. `rate_card` re-prices by platform × tier (batch is ~half);
# cache_read_tokens / cache_cost bill prompt-cache usage (see patterns 12–14).
from briefcase.cost import CostCalculator
calc = CostCalculator()
est = calc.estimate_cost("gpt-4o", input_tokens=1200, output_tokens=350, rate_card="batch")
print(est.total_cost, est.cache_cost)

Prefer the decorator? @capture records a function's inputs/outputs/timing automatically — call briefcase.observe("console") (or briefcase.setup(exporter=...)) once so the records are sent somewhere. See patterns/01_decision_capture.py.

See also: @capture decorator, CostCalculator, DriftCalculator, Sanitizer, GuardrailPipeline, ReplayEngine, BitemporalRecord, PolicyRegistry, AgentRouter, ExaminerBundle. Full SDK docs at briefcaseai.io.

Support

support@briefcaseai.org | briefcaseai.io | LICENSE

About

Sample use cases and workflows for using Briefcase AI

Resources

Code of conduct

Contributing

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors

Languages