TechLion

Process automation and agentic systems

I own the whole thing: mapping the process, designing the solution, building it, shipping it, and keeping it running.

I build solutions on large language models with deterministic steps and human approval wherever a model error could carry consequences.

Delivery practices

A person decides where mistakes are costly

The model gets the part of the job where it genuinely helps. The rest stays deterministic, and nothing reaches the target system without human approval.

Human in the Loop – the agent prepares a proposal; a person decides whether it is written.
Approval gates – approve, edit, or reject before every write to the target system.
Output validation – the model response is checked against its expected structure before it moves on.
Deterministic steps – whatever can be expressed as a rule stays a rule. The model is not responsible for work that needs no model.
Hallucination control – a narrowed scope of agent actions and answers grounded in data rather than guesswork.

Prompt strategy

Designing AI responses you can deploy

I stay focused on outcomes: AI responses should be predictable, consistent, and instantly usable in your processes. I design them to advance business goals, meet quality requirements, and plug smoothly into automations.

Context and data hygiene – only the essential context, clear sources, and refresh cadence.
LLM decision paths – when to rely on a model, on rules, or on user clarification.
“Ready-to-integrate” format – responses in JSON/Markdown with predictable structures for backend use.

Automation

Connecting tools and services

I design and launch workflows that help teams save time and remove manual work.

Integrations across no-code and low-code apps, over REST APIs and webhooks
AI in the loop: classification, extraction, summaries, and model-assisted decisions
Monitoring automation quality and security

Skills

Beyond the technology

Ownership of the whole delivery – from mapping the process to maintaining it
Non-obvious thinking and rapid prototyping
Streamlining everyday team processes
Facilitating workshops and transferring knowledge

Projects

What I have built

Automations, agentic systems, and prototypes I have shipped for clients and for myself.

Task agent working from meeting notes

Proposed tasks and comments land in a dedicated Slack channel: approve, edit in place, or reject in bulk. Only approved items reach the task tracker. n8n rather than an open agent framework – to keep the agent scope tight and force approval before every write.

n8n
Slack
Asana
Human in the Loop

Community support handled from Slack

Tickets and mentions reach the support team channel together with a link to the comment. Claiming a ticket and marking it done happens without leaving Slack.

n8n
Slack
Circle

Community conversation quality analysis

Posts and comments go through analysis for conversation quality, unresolved tickets, and moderation needs.

n8n
Airtable
Circle

Alerts for delegated tasks

Delegating a task or mentioning a user posts a message to a dedicated Slack channel, tagging the specific person.

n8n
Asana
Slack

OCR for handwritten documents

MVP

Digitising scans under GDPR constraints: processing stays local, with no data sent to external APIs. A working MVP in n8n on locally hosted models.

n8n
OCR
local models
GDPR

Agentic system running on local models

Personal project

Agent orchestration: work is delegated to specialised subagents. A conversation mode with speech synthesis and recognition. Runs on local models wherever data privacy matters, with an optional switch to hosted APIs.

subagents
local models
OpenRouter
OpenAI
Anthropic
speech synthesis (TTS)
speech recognition (STT)

I respect my clients' privacy and the individual character of every engagement, which is why I describe the solutions rather than the company names. Confidentiality is the default here, not an add-on – including where a separate agreement covers it.

AI Advantage