Evidence status is visible by design

Technical evidence and illustrative system patterns

A demonstration is not a client result. An architecture pattern is not a delivered project. This page keeps those categories separate until primary evidence and publication permission support a real case study.

Demonstration · repository-backed

Single-document Q&A prototype

Demonstrate a document upload, retrieval, and context-constrained answer flow using synthetic content.

Review prototype limitations

Architecture / work shown

  • Next.js upload and question interface
  • PDF or UTF-8 text extraction
  • 1,000-character chunks with 200-character overlap
  • TF-IDF/cosine lexical retrieval
  • Top-three chunks sent to Groq for generation
  • Temporary process-memory session store

Truthful result

A technical prototype existed in the repository. The public interactive path was retired after it failed the production audit and its data-handling/session limitations were identified.

Limitations

  • Not a client deployment
  • Not a semantic embedding or durable vector database
  • No verified citations, accuracy, latency, scale, privacy, or adoption result
  • No public upload is currently enabled

Illustrative pattern · synthetic data

Human-approved document workflow

Show how unstructured input could be extracted, validated, reviewed, and routed without allowing an unreviewed model to take a consequential action.

Architecture / work shown

  • Document or request intake
  • Extraction/classification candidate
  • Schema and business-rule validation
  • Human approve, correct, or reject step
  • Approved downstream API action
  • Failure log and manual fallback

Truthful result

This is a reusable architecture pattern, not a claim that a client system was deployed or produced a measured result.

Limitations

  • No client identity or adoption claim
  • No task, time, accuracy, or savings metric
  • Specific tools are selected only after assessment
  • A client case requires primary artifacts and permission

Illustrative pattern · synthetic data

Private knowledge-system pilot

Show the minimum complete path for testing answers over an approved source set before a broader internal knowledge deployment.

Architecture / work shown

  • Approved sources and access assumptions
  • Ingestion, parsing, chunking, and metadata
  • Retrieval candidate and answer layer
  • Source display and unsupported-answer state
  • Golden-set evaluation
  • Data-flow, retention, and handoff notes

Truthful result

This is an implementation pattern used to scope a pilot. It is not published as a completed client case.

Limitations

  • No blanket private, secure, compliant, or accurate claim
  • No corpus-size or performance metric
  • No client screenshot or document
  • Provider and storage choices remain assessment decisions

Publication gate

What a real case study must contain

Candidate client projects remain private until their work, ownership, measurements, assets, and exact public wording are verified. Projects are never combined to create a stronger-looking story.

  • A defined baseline and comparison period
  • A sample size and measurement definition
  • A reproducible report or benchmark
  • Environment, volume, and exclusions
  • An evidence owner and review date
  • Written publication permission when client work is involved

Have a workflow worth assessing?

Start with a non-confidential description, approximate volume, current systems, and the output a reviewer needs.

Request an assessment