Skip to content
Stefan Manja
EN СР
01 / selected work

Selected work

Delivery choices, system shape, and outcomes from real internal AI work.

These write-ups are intentionally selective: enough to show the problem shape, system approach, and proof of execution, without exposing internal material that should stay internal. Current public cases draw from in-house work at Delta Holding; external client work is added when disclosure allows.

How to read the proof

Read these cases by what changed, not by volume of claims.

Each case is public-safe and selective. The useful proof is the workflow shape, the constraint set, the operating choices and the measured signal.

Workflow changed

What process changed, who used the output and where human judgment stayed in the loop.

System shape

Why the work became a recommendation layer, internal assistant, hardening path or no-build decision instead of generic chatbot work.

Operating constraints

Access, review, ownership, cost visibility and handoff are part of the proof, not implementation detail.

Measured signal

Outcome or adoption is shown only where the measurement is public-safe.

Delta Holding

Case 01

Credit-risk decision-support workflow

A credit-risk decision-support workflow built in-house at Delta Holding. Combines public-company risk signals with internal context into recommendations.

Context
B2B credit-risk analysis and limit-setting workflow inside a multi-entity enterprise finance function.
Outcome
Observed workflow outcomes included ~75% faster analysis, 4.4/5 analyst-rated quality, and 90%+ recommendation acceptance.
  • Python
  • LLM APIs
  • Workflow design
  • Structured decision support
Read the case study →

Delta Holding

Case 02

Self-hosted internal knowledge assistant

A self-hosted internal knowledge assistant built in-house at Delta Holding, with Azure AD access, operator surfaces, and usage + cost observability.

Context
Internal enterprise knowledge-access system serving multiple business units, with requirements around access control, quality governance, and operational visibility.
Adoption and feedback signal
Deployed across roughly 1,500 eligible users, with 300+ actual users and 85%+ positive explicit feedback on thumbs-up/down responses.
  • Flask
  • Docker
  • PostgreSQL
  • pgvector
  • LangChain
  • Azure AD SSO
Read the case study →

Workflow review

If you have a similar internal AI workflow to build or harden, I can review the workflow shape.

A short note on the workflow, users, current stage, and constraints is usually enough to tell whether build or advisory work makes sense.

Send a workflow for review See workflow starting points