AI Systems & Prompt Engineering

What I do

AI, but with structure. I treat prompts like product requirements.

🧭

Model selection & routing

Pick the right model for the task (speed vs reasoning vs cost) and route requests accordingly.

multi‑model strategy
fallback behavior
cost control

🧪

Prompt engineering

Design prompts that produce consistent outputs: tone, structure, constraints, and edge cases.

system + developer prompts
format enforcement
iterative refinement

🛡️

Guardrails & reliability

Reduce hallucinations with templates, checks, and “ask‑back” behavior when info is missing.

input validation
refusal boundaries
safe outputs

Process

A simple workflow that keeps AI outputs consistent.

01

Define the output

What should the answer look like? Format, tone, length, and constraints.

02

Pick the model strategy

Fast model for drafting, strong model for reasoning, cheap model for cleanup—if needed.

03

Build guardrails

Templates, refusal rules, ask‑back questions, and “don’t guess” behavior.

04

Evaluate + iterate

Test with edge cases, improve prompts, and measure output quality.

Prompt Lab

Copy a clean prompt pattern and use it in your projects.

[ROLE]
You are a reliable AI assistant designed to produce accurate, structured, and decision-ready outputs.

[OBJECTIVE]
Transform unclear or incomplete user input into a clear, useful response that solves the underlying problem.

[PROCESS]
1. Identify the user’s intent.
2. Detect missing or ambiguous information.
3. Ask concise clarification questions if needed.
4. Generate the response using a clear structure.
5. Validate the answer for accuracy and usefulness.

[CONSTRAINTS]
- Do not assume missing information.
- Do not hallucinate facts.
- Prefer clarity over verbosity.
- Use neutral, professional tone unless specified otherwise.

[OUTPUT FORMAT]
- Brief summary
- Structured main response (steps / bullets / sections)
- Risks or limitations (if relevant)
- Clear next action for the user

Tip: change OUTPUT FORMAT to match your product UI.

Intent — What does the user actually want?
Missing info — What’s unclear or required?
Failure modes — What should never happen?
Format — What output is easiest to use?
Success — How will we measure good output?

I design AI workflows that turn messy inputs into clean, reliable outputs.

Use this with me

How the Router Thinks

Every prompt is evaluated, scored, and routed — not guessed.

User Prompt Raw input received

Intent Detection Task & domain classification

Complexity Scoring Reasoning depth & structure

Cost / Speed / Quality Matrix Constraint evaluation

Model Selected GPT-4 (optimal reasoning)

Live Decision Example

“Summarize a legal contract in simple terms”

Token LengthHigh

Reasoning DepthHigh

CreativityLow

Latency PriorityMedium

Cost SensitivityMedium

✓

Selected Model

GPT-4 — optimal reasoning accuracy

92% confidence

Want this for your project?

Send me your use case and I’ll suggest the best prompt + model setup.

Best for

websites, student bots, automation, content systems

Deliverables

prompt pack • model routing • evaluation checklist

Open main portfolio

Model Router