AGENT SYSTEMS / DATA · KNOWLEDGE BASE · WORK TASKS

Agent systems for working tasks and knowledge bases.

I set up agent systems for staff and managers: data collection, summaries, document search and working tasks. I connect RAG retrieval to find relevant knowledge-base passages with source references. The model can run locally or through an API. The anonymised scenarios below show workflows for a team at a hypothetical company.

My role
Assistant setup and working processes
What is inside
Information, a team knowledge base, materials and work tasks
Result
Start tasks and review results in Telegram.

What needed to be done.

An assistant uses selected sources and tools, while a team member reviews the result and makes decisions. Prospect research and outreach can use a separate process with fact checks and approval steps.

  • Set up assistants in Telegram, connected working tools and scheduled tasks.
  • Built research, selection, scoring and draft preparation steps.
  • I connect document processing, knowledge bases and development task delegation.
  • Added response checks and options to review, edit, reject or approve materials.

How it works.

  1. Receive a task and context

    Sources, notes or a task description enter the relevant workflow.

  2. Prepare a result

    Filters, material lookup and tools help build a shortlist, a note or a draft.

  3. Review and continue

    A person reviews the result, edits it or chooses the next step.

Assistants use the tools the task needs.

I used the existing Hermes platform and configured sources, schedules, tool connections and preparation processes.

Results can be checked and edited.

The system checks required response fields. A person decides what to do next; a failed step is shown in its status.

You can see where a task stopped.

Each stage has a status and an error record. A partial result is not presented as a completed task.

Implementation and example.

The examples illustrate implemented workflow types. Inputs and outputs are fictional; this explains their purpose and is not a recorded model run. English UI.

The full approval-to-send path needs a separate live check. These examples demonstrate preparation and review.

DIAGRAM OF THE IMPLEMENTED SYSTEM

  1. Sources, notes and tasks

    Information from configured sources and team documents.

  2. Agents and working tools

    Filtering, preparation, knowledge-base work and code delegation.

  3. A result for review

    A shortlist with links, a note or a draft to review.

Architecture diagram of the project. Live data and connections are not shown.

The result.

Staff can start research from Telegram, receive a shortlist or draft and review it. The scenarios show how staff can use assistants to work with sources, documents and tasks.

Technologies and implementation details

Hermes (NousResearch), Python, MCP, Codex, PostgreSQL, Pydantic, n8n, Gmail API and Telegram. The engine is ready-made; the agents, stages, checks and integrations are mine.

Let's discuss your task.

What recurring task takes up time for you or your team? It might be searching documents, collecting data or preparing a summary. We can review one example and decide which part AI could handle.

Discuss an agent system ↗︎
PROJECT SCREEN