Skip to content
DenseChip
Learn
PracticeCode labInterview

DenseChip

Navigate

Menu

LearnPracticeCode labInterview

Tools

All toolsSV editorCode repoResume
DenseChip

Practice, courses, and chip-design tools in one focused workspace.

LearnPracticeCode labInterviewToolsLearning paths

© 2026 DenseChip

AboutContactPrivacyTerms
Skip to lesson

Course contents

  • Getting Started with AI for VLSI Engineers5
  • Just Enough AI: Models and Context15
  • Prompting13
  • Trust and Verification14
  • Using AI for Engineering Work17
  • Engineering Walkthroughs7
  • Practical Foundations for AI Workflows4
  • Engineering Data5
  • Retrieval-Augmented Generation6
  • Agents and Tool Use18
  • Agent Design8
  • MCP and Skills6
  • Multi-Agent Design6
  • Practical Capstone6
  • Tokens, Cost and Latency12
  • Agentic Workflows for Chip Teams18
  • Interview and Judgment10
    • AI Interview Questions
    • Design Scenarios with Numbers
      • Estimate a Review Assistant Budget
      • Design a Regression-Triage Agent
      • Give a Model a 300k-Token Codebase
      • Evaluate Before Adopting a Tool
  • AI Skills: Full Course Index
  • Working with AI Glossary
  • Working with AI Cheatsheet
Loading lesson

‹Design Scenarios with Numbers
  1. Topics
  2. AI Skills for VLSI Engineers
  3. Interview and Judgment
  4. Design Scenarios with Numbers
  5. Estimate a Review Assistant Budget

Reference: Interview and Judgment · Beginner to Intermediate

Estimate a Review Assistant Budget

Estimate the monthly token cost of a code-review assistant from stated example parameters.

On this page

    Assumptions first

    No estimate holds without stated inputs. The table below fixes every input as an EXAMPLE parameter in fictional cost units.

    Parameter

    Example value

    Team

    8 engineers.

    Reviews

    2 per engineer per day.

    Days

    20 working days per month.

    Input per review

    6000 tokens of diff plus context.

    Output per review

    1200 tokens of comments.

    Input price

    3 units per million tokens (example, not a real price).

    Output price

    12 units per million tokens (example, not a real price).

    Reruns

    1 in 10 reviews runs twice.

    Token counts vary by tokenizer: measure them on real diffs. Prices vary by provider and over time: treat the units above as placeholders.

    Worked cost

    The program below checks the same arithmetic the calc block shows by hand. Both use only the table parameters.

    plaintext
    1
    2
    3
    4
    5
    6
    reviews per month = 8 x 2 x 20 = 320
    input tokens = 320 x 6000 = 1920000 = 1.92 million
    output tokens = 320 x 1200 = 384000 = 0.384 million
    base cost = 1.92 x 3 + 0.384 x 12 = 5.76 + 4.608 = 10.368 units
    with reruns: 10.368 x 1.1 = 11.4048 units, about 11.4 units/month
    input share = 5.76 / 10.368 = 55.6%, output share = 44.4%; 55.6 + 44.4 = 100.0

    Result: about 11.4 cost units per month under these examples. Recommendation: budget 12 units and re-measure after the first month.

    example.py
    1
    2
    3
    4
    5
    6
    7
    8
    9
    10
    11
    12
    13
    14
    15
    16
    17
    18
    19
    20
    21
    22
    23
    24
    25
    26
    27
    28
    29
    30
    import sys
    
    def monthly_cost(engineers, per_day, days, in_tok, out_tok, in_price, out_price, num, den):
        reviews = engineers * per_day * days
        in_tokens = reviews * in_tok
        out_tokens = reviews * out_tok
        base_mu = (in_tokens * in_price) // 1000 + (out_tokens * out_price) // 1000
        total = base_mu * num / den / 1000
        return reviews, in_tokens, out_tokens, base_mu, total
    
    def main():
        reviews, in_tokens, out_tokens, base_mu, total = monthly_cost(8, 2, 20, 6000, 1200, 3, 12, 11, 10)
        if reviews != 320:
            print("FAIL: reviews total is off")
            sys.exit(1)
        if in_tokens != 1920000:
            print("FAIL: input tokens total is off")
            sys.exit(1)
        if out_tokens != 384000:
            print("FAIL: output tokens total is off")
            sys.exit(1)
        if base_mu != 10368:
            print("FAIL: base cost milli-units off")
            sys.exit(1)
        if "%0.2f" % total != "11.40":
            print("FAIL: monthly total rounding off")
            sys.exit(1)
        print("PASS: 320 reviews cost 11.40 units with 10 percent reruns")
    
    main()

    What changes the answer

    Volume scales the total linearly. The input and output price ratio decides whether shorter replies matter.

    If this changes

    Monthly total becomes

    Diff size doubles to 12000 input tokens

    About 17.7 units.

    Reviews rise to 4 per engineer per day

    About 22.8 units.

    Output price rises to 24 per million

    About 16.5 units.

    Common mistakes

    Common pitfall

    Counting only input tokens understates the bill. Output tokens often carry the higher unit price, and reruns add about a tenth here.

    Interview

    Likely probes: the estimate structure and its weak points.

    • Start from reviews per month, then tokens per review, then unit prices.

    • Token counts are measured on real diffs since they vary by tokenizer.

    • Reruns and the output price ratio belong in the total, not as footnotes.

    Check yourself Reviews rise from 2 to 4 per engineer per day with all else fixed. What is the new monthly total?

    Reviews double to 640, so the total doubles to about 22.8 units. Every term scales with review count.

    Check yourself Output price doubles to 24 with all else fixed. What is the new total?

    Output cost becomes 0.384 x 24 = 9.216. Base is 5.76 + 9.216 = 14.976, times 1.1 is 16.4736, about 16.5 units.

    Related topics

    • Cost arithmetic

    • Reducing tokens

    Previous

    167 / 173

    Next: Design a Regression-Triage Agent

    Up next

    Up nextDesign a Regression-Triage Agent

    Sign in to save progress

    Practice this lessonReport an error

    See also

    • Give a Model a 300k-Token Codebase
    • Evaluate Before Adopting a Tool