Applied AI systems

No “coming soon” claims · working repositories only

The LLM is one component. The system is the work.

These projects focus on orchestration, validation, retrieval, interfaces and failure boundaries — the layers between a promising model call and a usable product.

01 / Live research demo

Gemma 4 Hackathon · 2026

ImciFlow

LLM extraction, RAG evidence and deterministic IMCI rules in one auditable workflow.

A multilingual clinical decision-support prototype that combines Gemma 4, retrieval over IMCI references, deterministic safety rules and a human-review boundary.

3

languages

English, French and Sudanese Arabic paths

Gemma 4LangGraphFastAPIReactChroma
The evaluation does not yet measure end-to-end extraction quality, multilingual robustness or clinical outcomes. The hosted demo is research software and must not be used for diagnosis.
Evidence case study
01020304PROFILE → RECOMMEND → VALIDATE → APPROVE
02 / Tested prototype

Applied AI engineering · 2026

GenAI Data Prep

Deterministic data-quality checks first; LLM recommendations second.

A LangGraph workflow that profiles a dataset, proposes preprocessing decisions, validates transformations and keeps a human approval point before output generation.

81/81

tests

Verified locally with an isolated test configuration

LangGraphFastAPIStreamlitPydanticOpenAI
Data samples can be sent to an external model when that provider is enabled. Production use requires explicit consent, redaction, path/URL restrictions and a retention policy.
Evidence case study
Engineering boundary

What sits around the model.

A reliable AI system needs more than an orchestration diagram. These are the controls currently demonstrated in the public work.

01

Structured output

Pydantic contracts reject malformed or out-of-range model responses.

02

Deterministic controls

Statistical and safety rules remain testable outside the model path.

03

Human authority

Approval and final decisions stay with the user in high-impact workflows.

04

Evaluation scope

Rule tests, retrieval quality and end-to-end model behavior are reported separately.

Demonstrated now

  • FastAPI services and React/Streamlit interfaces
  • LangGraph orchestration and retrieval pipelines
  • Schema validation, repair loops and deterministic tests
  • Vercel and Cloud Run deployment

Still being strengthened

  • End-to-end multilingual and retrieval evaluation
  • Authentication, rate limiting and data-retention controls
  • Production observability for quality, latency and cost
  • Provider-independent test isolation
Browse the full evidence library