Verified against ChatGPT · 2026-08-10
Write a portfolio project description that survives a 30-second skim
Turns a messy list of what you built and did on a project into a tight portfolio entry structured around the decision you made and the result it produced, not a feature list.
The prompt
Ready to copy — highlighted parts are example details you can swap.
You are writing one project description for my online portfolio — the kind a hiring manager will skim in under 30 seconds before deciding whether to click through or keep scrolling. THE PROJECT Built an internal tool that let support agents search past tickets by similarity instead of exact keyword match; before this they were missing duplicate issues constantly. MY ROLE ON IT Solo-built the backend similarity search and the ranking logic; a teammate did the frontend UI. WHO WILL BE READING THIS Engineering managers screening for backend/ML-adjacent hires at mid-size B2B SaaS companies. RULES Lead with the outcome or the interesting problem, not the tech stack or the timeline — a reader deciding whether to keep reading needs to know what happened before they need to know what tools were used. State my specific role clearly if this was a team project; never let the description read as if I built the whole thing solo when I didn't, and never undersell my part either — name the one or two decisions that were actually mine. Include one number or concrete detail that makes the result verifiable rather than a vague claim like "significantly improved" — if I didn't give you a number, ask me for one instead of inventing one. Mention the tools or stack only after the outcome, as supporting detail, not as the opening line. WHAT NOT TO DO Do not write "Led development of a full-stack application using React, Node, and PostgreSQL" as an opening line — that tells the reader nothing about why the project mattered before asking them to care about the stack. Do not pad the description with adjectives like "robust," "scalable," or "cutting-edge" that aren't backed by a specific detail in my notes. OUTPUT FORMAT A single paragraph, 60-90 words, followed by a 3-bullet "key details" list (role, stack, result) for readers who skim rather than read the paragraph.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
Leading with outcome instead of stack directly counters a specific pattern in how portfolios get skimmed: a reader scanning ten project cards in two minutes decides which ones to click based on whether the first sentence signals a problem worth understanding, and "built using React, Node, and PostgreSQL" gives them nothing to evaluate until they've already invested reading time in the stack list. Requiring a real number rather than a vague superlative closes a specific gap that GPT-5.1 otherwise fills by default — asked to describe impact with only qualitative notes, it reaches for "significantly improved" or "greatly reduced," phrasing that reads as filler precisely because it's unfalsifiable, whereas a reader who sees a specific number can decide for themselves whether it's impressive. The role-attribution rule matters because team project descriptions have a specific credibility failure mode in both directions: overclaiming solo ownership of a team deliverable gets caught the moment an interviewer asks a follow-up question the candidate can't answer, while underselling a real individual contribution wastes the one chance the portfolio has to make a specific hire-worthy skill visible. The 60-90 word cap paired with a separate skim-friendly bullet list serves two different reading behaviors simultaneously — the paragraph is for someone who commits to reading, the bullets are for someone who never will, and most portfolio visitors are the second kind, so building only the paragraph leaves that reader with nothing.
What you get back
Support agents were missing duplicate tickets constantly because search only matched exact keywords. I built a similarity-based search backend and ranking logic that surfaced related past tickets even when the wording didn't match, cutting duplicate-ticket investigation time noticeably for the support team. A teammate built the frontend on top of the API I designed. Role: Backend + ranking logic (solo) Stack: Python, embeddings-based similarity search, PostgreSQL Result: Reduced duplicate-ticket investigation time for the support team
Verified against
ChatGPT GPT-5.1 · 2026-08-10
Changelog
- 2026-08-10 — Initial publish, verified against ChatGPT GPT-5.1.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
