(00) About Job Atlas

Why this exists

A job search is notone problem. It is six,pretending to be one.

Boards, spreadsheets, resume tools, referral threads, and a calendar that nobody updates. Each one holds a piece of the truth and none of them talk. Atlas is the assumption that the pieces were always one system, and that treating them that way is what finally makes the search legible.

Launch market: India, across boards, startups, and internships
Inputs: Six live signal sources, versioned and timestamped
Scores: Candidate, company, and fit, computed separately
Ranking: Deterministic weights today, outcome-learned next

Youalreadyknowwhichpartsofyoursearcharebroken.Whatyoudonothaveisasystemthatagreeswithyouinwriting,ranksthedamage,andtellsyouwhichrepairpaysforitselffirst.

(01) Principles

What gets built, and what does not

Six decisions that
settle every argument.

Productdebatesarecheapwhentheprinciplesarewrittendown.ThesearetheonesAtlaskeepsreturningto.

Every score has to open.

Explainability

A number you cannot argue with is a number you cannot act on. Each score in Atlas decomposes into the signals that produced it, which means a low one is not a verdict. It is a list of things to go fix, in order of how much they cost you.

India first, but never India only.

Scope

Launch scope is the US, India, and remote-friendly UK/EU roles. Sources, regions, and providers sit behind interfaces from day one, so the next market is a configuration change rather than a rewrite.

Language models enrich. They do not rank.

Architecture

LLM work belongs in parsing, drafting, and enrichment, where a second pass can catch a bad answer. Ranking runs on deterministic features with versioned weights, so the same profile and the same posting produce the same score twice.

Automation assists. It never hides.

Control

Applications, outreach, and follow-ups stay editable, reversible, and visible. Nothing is sent that you did not see. The moment a product starts acting on your behalf in the dark is the moment you stop trusting the parts that work.

Outcomes are the only honest teacher.

Learning

Replies, interviews, offers, and rejections are captured against the exact feature snapshot that existed when the recommendation was made. That is what lets ranking improve from evidence instead of from opinion.

Private context stays private.

Data

Learning happens on structured signals, not on the contents of your inbox or the text of your conversations. The system gets better at the market without needing to read your life.

(02) The stack

Data, scoring, control

Three layers, and
a hard line between them.

(01) Data

Six inputs, one shape.

Postings, company records, resumes, code evidence, professional profiles, and contacts are parsed into versioned features with a source and a timestamp on every one. If a number moved, you can find the row that moved it.

(02) Scoring

Three scores, kept apart.

Candidate strength, company quality, and match fit are computed independently. Bundling them into one figure is how products end up unable to explain themselves, and how candidates end up improving the wrong thing.

(03) Control

Manual override, always.

Every automated step has a hand-operated version sitting next to it. Atlas is meant to remove the tab management and the spreadsheet, not the judgement.

(06) Who it is for

Built for people who need a repeatable system.

Atlas works best when the search has many moving parts: roles, documents, people, deadlines, and follow-ups. It keeps them connected instead of scattered.

Students & fresh graduates

Find internships and first roles, build real resume evidence, track cold emails, and learn which profile gaps matter most right now.

Professionals changing roles

Prioritise realistic opportunities, compare companies, tailor resumes, and keep interviews, assessments, and follow-ups in one place.

Referral-led searches

Connect people, companies, role context, and outreach history, without turning your job search into a sales pipeline.