Daynis OlmanAI, Cloud & Digital Platform Leader

Grant Application Intelligence Lab

A prototype for evidence-aware grant application review: document ingestion, classification, retrieval, criterion-level analysis, consistency checks, calibration and human-readable reporting.

Evidence first, then judgement.

  1. 1. Application document
  2. 2. Evidence retrieval
  3. 3. Criterion review
  4. 4. Human-readable report
Evidence: Innovation prototypeAI product design
Role
Product concept and prototype engineering
Serves
Reviewers assessing applications against defined criteria.
Core technologies
RAG · Document AI · Evaluation

Executive overview

The problem. Reviewing grant applications against criteria is slow and can be inconsistent between reviewers. This prototype explores how AI could support, not replace, evidence-based review.

Who it serves. Reviewers assessing applications against defined criteria.

What Daynis did. Product concept and prototype engineering. A prototype for evidence-aware grant application review: document ingestion, classification, retrieval, criterion-level analysis, consistency checks, calibration and human-readable reporting.

Why it matters. Explores how evidence-aware AI can make review more consistent and transparent while keeping reviewers in control. This is an innovation prototype, not a deployed system.

Business architecture and impact

Capabilities created

  • Ingest and classify application documents
  • Analyse evidence criterion by criterion
  • Consistency checks and calibration
  • Human-readable review reports

Explores how evidence-aware AI can make review more consistent and transparent while keeping reviewers in control. This is an innovation prototype, not a deployed system.

System architecture

  1. Reviewers + Application documents

    Where requests and source data originate.

  2. Review report

    The interface people use day to day.

  3. Ingestion & classification

    Orchestrates requests and enforces business rules.

  4. Criterion-level analysis

    Models applied under defined, reviewable constraints.

  5. Evidence index

    The governed system of record.

Governance: Consistency checks & calibration
Documents are ingested and classified, then indexed for retrieval. For each criterion, relevant evidence is retrieved and analysed. Consistency checks and calibration compare findings, and results are compiled into a human-readable report for the reviewer.

Key flows

  • Application documents to Ingestion & classification
  • Ingestion & classification to Evidence index
  • Evidence index to Criterion-level analysis
  • Criterion-level analysis to Consistency checks & calibration
  • Consistency checks & calibration to Review report

Technical depth

Highlights

  • Document ingestion and classification
  • Retrieval of supporting evidence
  • Criterion-level analysis
  • Consistency checks and calibration

Architecture decisions

Evidence before verdicts

Every finding is tied to retrieved evidence a reviewer can inspect.

Calibrate, don't assume

Consistency checks surface disagreement rather than hiding it.

Security and governance

  • Human reviewers remain the decision-makers

What I led

  • Product concept and prototype architecture

Evidence and links

Evidence: Innovation prototypeThis is an innovation prototype and has not been deployed as a production system.

Building something similar?

Talk to Daynis about ai product design architecture and delivery.

Start a conversation