AI engineering for public institutions

Structured AI prototyping for municipal, governmental, and budget-funded organizations.

Possible areas of application in these organizations.

Scenario groups

Preliminary document triage and draft summaries

Documents are sorted and summarized as drafts; a responsible employee decides.

  • Preliminary sorting of documents
  • Draft summaries for an employee to check
  • Employees remain responsible for decisions

Processing text collections

Larger sets of texts are processed in a bounded way to find patterns or prepare drafts.

  • Grouping similar texts
  • Preparing draft overviews
  • Checking results on a sample first

Internal knowledge assistance

A developing RAG direction for questions about internal documents. It is not an available service.

  • Questions answered from internal documents
  • Sources shown for checking
  • Developing direction, not a service

Possible solution scenarios

Concrete organizational problems we can build solutions for.

Further scenarios we can build on top of the directions above.

Internal AI knowledge assistant

Developing

A private assistant for organizational documentation: semantic search and answers over approved documents, with links to the sources. Role-based access belongs to the target architecture only where it is specified and built.

Possible workflow

  1. Approved documents
  2. Indexing
  3. Semantic search
  4. Answer with source links
  5. Human check

Boundary: Answers from a knowledge base work in the Telegram secretary, and document stores with search in OpenAI API Desk. Role-based access and private deployment are not claimed.

More projects: Electronic Telegram secretary, OpenAI API Desk

Electronic reception and intelligent request routing

Confirmed by projects

A website, Telegram or another permitted channel receives a request. A workflow can understand it, classify the topic, assess urgency, identify the responsible department, create a structured request, route it and track its status. The person does not need to know which department or specialist should receive it.

Possible workflow

  1. Request received
  2. Topic and urgency
  3. Responsible department
  4. Structured request
  5. Routing
  6. Status tracking, reminders, escalation

Boundary: Routing by topic, saving the request and notifying the responsible employee work in the Telegram secretary and the feedback form. Statuses, priorities, repeated notifications and escalation are the next step.

More projects: Electronic Telegram secretary, Feedback form with automatic routing

Intelligent incoming-document routing

Demonstration concept

Incoming email or documents are read, classified and routed to the responsible department, with a task created for a person to review. Document types could include applications, requests, official letters, service notes, complaints and commercial inquiries.

Possible workflow

  1. Email or incoming document
  2. Text extraction
  3. Classification
  4. Metadata extraction
  5. Responsible department
  6. Routing and task creation
  7. Human review

Boundary: No legal compliance, automatic official registration or legally valid document processing is implied. A person reviews every routing result.

AI-assisted document preparation

Demonstration concept

Drafts of responses to incoming requests, service notes, protocols, summaries, reports, letters and structured case summaries, prepared for a person to review and approve.

Possible workflow

  1. Source material
  2. Draft
  3. Human edit
  4. Human approval

Boundary: Drafts only. Responsibility stays with the organization and its people.

AI analysis of organizational correspondence

Demonstration concept

Accumulated requests and correspondence are analyzed to show recurring topics, workload by department, repeated questions, unresolved requests, response times and places where automation may be useful.

Possible workflow

  1. Accumulated requests and correspondence
  2. Topics
  3. Workload by department
  4. Repeated and unresolved requests
  5. Response times
  6. Automation candidates

Boundary: An analytical use case that informs people. It is not an autonomous decision-making system.

AI document consistency review

Demonstration concept

Documents are checked for missing fields, conflicting dates, inconsistent names or identifiers, missing required sections, deviations from a supplied template and contradictory information.

Possible workflow

  1. Document and attachments
  2. Field and section checks
  3. Cross-document comparison
  4. Discrepancy flagged
  5. Human review

Boundary: It flags possible issues and a person decides. It is not a legal or compliance check.

All possible scenarios

What we would measure

Suggested measures for the first stage.

  • Time to a reviewed draft
  • Completeness of summaries against a sample
  • Correctness found on review
  • Manual corrections
  • Employee acceptance of drafts

First-pilot requirements

  1. Define which data is permitted
  2. Identify where processing happens
  3. Agree access
  4. Preserve employee responsibility
  5. Plan error handling and logging
  6. Start with a noncritical process

Constraints

  • Nothing here implies procurement eligibility, legal compliance, certification or completed institutional deployments.
  • Employee responsibility for decisions is preserved.

Applicable current capabilities

Confirmed by our projects.

  • Document processing Confirmed by projects
  • Local LLM experimentation Confirmed by projects
  • Model comparison Confirmed by projects

Developing directions

Directions under development.

  • Retrieval / RAG Developing

Relevant engineering projects

Requests & routing · AI assistants & Telegram bots

Electronic Telegram secretary

Takes over the first line of communication — from answering a question to passing a request to the right employee.

An AI secretary answers typical questions from a knowledge base and passes requests that need staff involvement to employees.

Requests & routing · Process automation & integrations

Feedback form with automatic routing

Requests from the website go straight to the right employee, without manual forwarding.

The form does not just collect messages — it takes over their initial processing and distribution.

First-stage deliverables

  • Solution diagram
  • A bounded working scenario
  • Test examples
  • A report with known limitations
  • A recommendation to continue, revise or stop

Start with one process.

Describe what is currently done manually, what data is involved and what result you would like to achieve. You do not need a technical specification or a chosen model to begin.

Fill in the brief
AI bot consultation (opens Telegram, an external service)

Please do not share trade secrets, source code, or sensitive personal data before an NDA (non-disclosure agreement) is signed.