HackerRank open sourced its ATS. My resume scored 90/100. Oh wait 74. No – 88

📝 Discussion Summary (Click to expand)

Key Themes

AI screening is arbitrary and can discard qualified candidates.

"I fail 65% of the time. Same exact resume, different luck." — jerrythegerbil
LLM evaluation claims are flawed, especially regarding deterministic behavior.

"temperature 0 — low, supposedly nudging the model toward deterministic outputs" — dvt
High application volume forces aggressive filtering, but current methods select the wrong signals.

"They are selecting for people who are fine working in their free time. If you contribute to open source you are more likely to contribute to the company on weekends. If instead you have other hobbies or a family that takes up non‑work hours you are more likely to drop your pen after forty hours." — adrianN

🚀 Project Ideas

Generating project ideas…

Solves the arbitrary “percentage‑score” black‑box that randomly drops 65% of applicants and instead gives a clear pairwise ranking.
Core value: lets recruiters see who is relatively stronger among candidates without relying on an opaque single score.

Key	Value
Target Audience	HR tech startups, recruiting platforms, remote hiring teams
Core Feature	Pairwise LLM comparison engine with deterministic output and traceable rationale
Tech Stack	Python, LangChain, Gemma 3 4B (or GPT‑4‑lite), SQLite for logging
Difficulty	Medium
Monetization	Revenue-ready: $0.01 per comparison tier

Appeals to the “compare‑pairs” idea raised by many commenters (“compare every pair of CVs for best results”, “take an elo system”) and directly addresses jerrythegerbil’s 35 % chance comment.
Low barrier to integrate into existing ATS pipelines, giving users transparent “why this candidate scores higher” explanations.

Removes or masks demographic cues (name, gender, graduation year, address) before feeding a resume to any LLM scorer, preventing the biased ranking highlighted by several users.
Core value: gives fairer, legally safer resume scores while still benefiting from AI‑driven screening.

Key	Value
Target Audience	Job boards, freelance platforms, individual job seekers
Core Feature	Automatic demographic masking + alternative keyword‑based scoring fallback
Tech Stack	Node.js, OpenNLP for entity detection, GPT‑3.5 summarizer, PostgreSQL
Difficulty	Low
Monetization	Hobby

Direct response to the concern raised by “Blue‑collar” and “sph” that GPA, college, and demographics unduly affect scores.
HN commenters lamented the “illegal bias” and “unfair filtering”; this tool turns that pain point into a sellable service.

Helps candidates craft resumes that highlight concrete skill evidence (e.g., commit‑message alignment, project‑specific keywords) based on LLM‑validated criteria, reducing reliance on luck.
Core value: increases the probability of passing AI filters by ensuring the resume matches what the LLM expects, rather than random resume luck.

Key	Value
Target Audience	Job seekers, career coaches, freelancers
Core Feature	Guided resume editor that auto‑generates skill‑rich bullet points and creates multiple stylistic variations to test LLM scoring
Tech Stack	React front‑end, Flask backend, Llama 3 8B for extraction, JWT auth
Difficulty	Medium
Monetization	Revenue-ready: $5/month subscription

Echoes the “multiple applications with slight wording variations” idea discussed by users like IshKebab’s “different names” hack.
Directly tackles the “random discard” frustration mentioned by many commenters who feel “unlucky” resumes never get reviewed.