50 mins lesson duration•10 mins read

Lesson 21: Tree of Thought & Persona Prompting

Explore several reasoning branches before deciding, and assign the model a persona to unlock specialized perspectives — with examples.

Two Techniques for Harder Problems

1. Tree of Thought (ToT)

Chain-of-thought walks one path. Tree of thought walks several branches in parallel, evaluates each, and picks the best — like a chess player considering moves.

When to use: strategy, planning, multi-criteria decisions, anything with trade-offs.

Example — choosing a marketing strategy:

I run a startup with a small budget. Evaluate 3 strategies
and pick one:

Strategy A: Influencer marketing
  - Cost: high | Time to results: medium | Risk: medium
Strategy B: Content marketing
  - Cost: low | Time to results: slow | Risk: low
Strategy C: Paid ads
  - Cost: medium | Time to results: fast | Risk: high

For each: score it 1-5 on cost, speed, and risk.
Then recommend the best for a startup with limited budget.

The model explores each branch, then compares scores instead of committing to one path early.

Prompt pattern for ToT:

1. List N candidate approaches.
2. For each, list its pros, cons, and risks.
3. Score each on [criteria].
4. Recommend the best and explain why.

2. Persona Prompting

Assign the model a role with expertise, perspective, and constraints. Personas change what the model attends to.

Persona	Unlocks
"Senior tax consultant"	Compliance-focused, precise language
"10-year-old explaining to a friend"	Simple analogies, no jargon
"Skeptical code reviewer"	Security and edge cases
"Hiring manager"	Candidate-focused feedback

Weak vs. strong persona example:

Weak: "Review my resume."
Strong: "You are a senior engineering hiring manager at a
startup. Review my resume for a Senior Backend role. Focus on:
impact metrics, missing keywords, and red flags. Give 5
actionable fixes ranked by impact."

Persona + ToT Combined

The power move: assign each branch a different persona, then synthesize:

Should we launch in Q1 or Q2?
- Analyze as the CFO (costs and cash flow).
- Analyze as the CMO (market timing and demand).
- Analyze as the Head of Engineering (team readiness).
Then give one recommendation weighing all three.

Common Mistakes

Personas that are too vague ("be an expert") — name the specific expertise and goal.
ToT without a scoring rule — branches need criteria to compare.
Asking for a persona but not honoring its constraints in the output format.

Key Takeaways

Tree of thought explores branches before deciding; score each branch.
Personas steer attention — the more specific the role, the better the result.
Combine: persona-per-branch + a final synthesis step.
ToT costs more tokens — reserve it for real decisions.

Next up: Prompt chaining — breaking big jobs into a sequence of focused prompts.

Interactive Lesson Code Snippet

# Tree of thought: evaluate branches, pick the best
branches = {
    "A: influencer marketing": {"cost": "high", "risk": "medium", "fit": "strong"},
    "B: content marketing": {"cost": "low", "risk": "low", "fit": "strong"},
    "C: paid ads": {"cost": "medium", "risk": "high", "fit": "medium"},
}

print("Branch evaluation (score = count of 'low'/'strong' ratings):\n")
for branch, metrics in branches.items():
    score = sum(1 for v in metrics.values() if v in ("low", "strong"))
    bar = "#" * score
    print(f"  {branch:30s} score={score} {bar}")

print("\nBest branch: B: content marketing")

Language: python

Lesson Code (Python)

# Tree of thought: evaluate branches, pick the best
branches = {
    "A: influencer marketing": {"cost": "high", "risk": "medium", "fit": "strong"},
    "B: content marketing": {"cost": "low", "risk": "low", "fit": "strong"},
    "C: paid ads": {"cost": "medium", "risk": "high", "fit": "medium"},
}

print("Branch evaluation (score = count of 'low'/'strong' ratings):\n")
for branch, metrics in branches.items():
    score = sum(1 for v in metrics.values() if v in ("low", "strong"))
    bar = "#" * score
    print(f"  {branch:30s} score={score} {bar}")

print("\nBest branch: B: content marketing")

Console Output

Branch evaluation (score = count of 'low'/'strong' ratings):

  A: influencer marketing        score=1 #
  B: content marketing           score=3 ###
  C: paid ads                    score=0

Best branch: B: content marketing

Code Visualization Tips

Draw the decision as a tree: root question → 3 branches → scores → chosen branch highlighted.
Label each persona in a group discussion with its priorities (CFO = cost, CMO = timing…).
Sketch the persona dial: same prompt, different lens — show what each lens notices.

Professional Tips & Tricks

Give every ToT branch an explicit scoring rule, or the comparison is vibes.
Name the persona's constraints ('you must refuse unsafe requests') to keep it honest.
For big decisions, one prompt per persona, then a final synthesis prompt — cleaner than one mega-prompt.

Python Code Judge & Practice Arena

LeetCode Style

Run real Python 3.12 WebAssembly code directly in your browser against automated test suites.

Solved:0 / 1

0 / 20 XP

Challenges:

Problem 1 of 1

Persona × 3 Decision

Medium+20 XP

Write a tree-of-thought prompt that evaluates 'hire a freelancer vs. hire full-time vs. use an agency' using three personas (CFO, Head of Product, Founder) and a scoring step.

main.pyPython 3.12 (WASM)

# Persona × 3 Decision
# Write your solution below

def solve():
    pass

Press Run Code to test or Submit to verify test cases

Test Your Knowledge

Instant feedback

Quick Check: Tree of Thought & Personas

1 / 2

How is tree of thought different from chain-of-thought?

Up next · Continue learning

Prompt Chaining & Multi-Step Workflows

Break big tasks into focused prompts where each output feeds the next — with a full worked chain and validation checkpoints.

10 mins read55 mins

Start next lesson

Previous: Self-Consistency — Sample & Vote Next: Prompt Chaining & Multi-Step Workflows

Made withbyAmol Shukla·amolshukla.online