b4joinacompany

Research a company before you apply, interview, or accept an offer.

b4joinacompany is a decision-support tool for job seekers in Bangladesh. It turns workplace stories, community-submitted salary information, reported work arrangements, and official company links into an evidence-labeled company brief. The goal is not to declare a company “good” or “bad”; it is to help someone understand what has been reported, notice what is still unknown, and prepare better questions before joining.

The product is built around a simple principle: personal reports are useful context, but they are not verified company facts. Every summary remains traceable to its source, uncertainty stays visible, and missing evidence is shown as a gap rather than treated as a negative signal.

Research a company · Review the methodology

Project status: the public research experience is usable without an account. Google sign-in is required for private saved checkpoints, and the optional AI-backed Ask flow depends on configured provider quotas.

Why b4joinacompany?

Job seekers often have to assemble a decision from anonymous posts, salary submissions, company pages, and interview conversations. Those sources differ in reliability and are easy to flatten into an unfair “good/bad company” judgment. b4joinacompany keeps provenance and uncertainty visible so the reader can prepare better questions without mistaking reports for verified policy.

What you can do

• Research a company — search confirmed company records and open a single brief covering culture, pay, work setup, source stories, and questions to ask.
• Prepare for a decision — get company-specific prompts for an application, interview, or offer discussion from a deterministic, rules-based question engine.
• Ask the evidence — ask a focused question and receive a cited answer from relevant workplace-story excerpts, or an explicit evidence gap.
• Compare two companies — review the same evidence categories side by side, including salary for the same role, without a synthetic score or winner.
• Save research privately — sign in with Google to keep immutable company checkpoints and the evidence revision used for each one.
• Research while browsing — use the companion browser extension on Deshi Mula company and story pages without creating a b4joinacompany account.
• Submit corrections — flag company or evidence issues for manual review; submissions never alter the published dataset automatically.

Using the application

Research one company

1. Open the company search.
2. Select a confirmed company record rather than relying on a name-only guess.
3. Review the evidence timestamp, source labels, workplace stories, reported work arrangement, salary evidence, and official destinations.
4. Open original source links for any claim that matters to your decision.
5. Use the suggested questions during the application, interview, or offer discussion to verify what is current.

Compare or ask

1. Open Compare and choose two companies to inspect the same evidence categories side by side; the product does not choose a winner.
2. Open Ask, enter one focused question, and check each [S1], [S2], or later citation against the retrieved story context.
3. Treat an evidence gap as “unknown,” not as evidence against the company.

Save a checkpoint or submit a correction

Sign in with Google to save a private snapshot of a company and its dataset revision. Use the correction form when an identity, destination, or evidence record needs review; a maintainer must approve any dataset change.

How evidence is handled

b4joinacompany keeps different kinds of information separate:

Label	Meaning
Reported	Personal workplace experiences with links back to the original story
Submitted	Community-provided salary context, including sample size and missing pay-period information
Derived	Deterministic interpretations such as reported work setup; never presented as current company policy
Official	Accepted company destinations such as a website, careers page, or LinkedIn profile

Sources and references

The main workplace source is Deshi Mula. Company pages, workplace stories, and comments are captured in a versioned local dataset. Role-based salary ranges come separately from Beton Kemon community aggregates. The app does not present either source as company-verified data.

Information shown	How it is selected or produced	Reference shown to the user
Company identity	A Deshi Mula company record supplies the canonical slug, source name, and snapshot date. Search also checks accepted display names, aliases, website URLs, and LinkedIn URLs.	The brief links to the original Deshi Mula company page and displays its evidence snapshot.
Related workplace posts	A post is related to a company only when its captured company_url ends with that company's canonical /companies/:slug path. Posts are not attached through keyword guessing.	Every public post keeps its Deshi Mula source_url, title, reported role, and publication date so the original story can be opened.
Comments and work setup	Comments are joined to their parent post by story_id. Reported remote, onsite, hybrid, schedule, and overtime mentions are deterministic derivatives of those posts and comments.	Each work-setup evidence mention retains its source kind, source ID, parent story ID, excerpt, role, date, and original story URL.
Salary information	Beton Kemon role ranges are joined to the company by an exact canonical-name match or a recorded manual-name review. Records retain the raw BDT range, sample size when available, capture time, and match confidence.	The salary section names Beton Kemon, shows the capture date and match method, and links to the original Beton Kemon company page.
Suggested questions	A fixed rules engine scans the related story bodies and joined comments for recurring terms across management, pay, stability, growth, workload, culture and safety, workplace flexibility, and hiring. Matching themes are ranked; they are used to form questions to verify, not findings about the company.	Each suggestion explains why it was raised and cites up to three matching stories or comments. Story references include title, reported role, and date; comment references include date and comment ID. If evidence is thin, the question is marked as an evidence gap.
Ask answers	The app retrieves a bounded set of company posts relevant to the user's question. Gemini or Groq may compose an answer; without either provider, the app returns a deterministic summary.	Claims use [S1], [S2], and similar labels that map to the retrieved story's role and date. The original posts remain available in the company brief.

The corresponding release records live in data/companies.jsonl, data/stories.jsonl, data/comments.jsonl, and data/company_salary_evidence.jsonl. These files preserve source URLs and matching metadata so the displayed information can be audited back to its origin.

Prepared checkpoint questions are generated locally by fixed rules and do not send report text to an AI provider. The optional Ask feature sends only a bounded set of relevant excerpts to the configured provider and requires cited answers. If no provider is configured, it returns a deterministic, source-labeled fallback.

The app does not:

• fact-check anonymous workplace claims;
• turn missing salary or hiring evidence into a negative signal;
• infer an onsite policy when work-arrangement evidence is unknown;
• combine unlike evidence into a company rating; or
• scrape, edit, or publish the source dataset through the web interface.

Tech stack

• Next.js 16, React 19, and TypeScript
• Tailwind CSS
• Better Auth with Google sign-in
• MongoDB for authentication, private workspaces, quotas, and metadata-only generation logs
• Gemini and Groq with a deterministic fallback for evidence-based answers
• A versioned, locally processed Deshi Mula dataset

How the application fits together

flowchart LR
  Sources[Versioned source datasets] --> Research[Deterministic research layer]
  Research --> Web[Next.js web experience]
  Research --> API[Versioned API]
  Web --> Ask[Optional Gemini / Groq answer synthesis]
  Web --> Workspace[(Private MongoDB workspace)]
  API --> Extension[Companion browser extension]

Loading

The versioned JSONL files are the public evidence base. Deterministic matching and question generation run before any optional AI synthesis. MongoDB is used for authentication, private checkpoints, correction submissions, quotas, and operational metadata—not as the source of published workplace claims.

Run locally

Requirements:

• Node.js 20.9 or newer
• pnpm 11
• MongoDB if you want authentication, saved workspaces, or persistent AI quotas

git clone https://github.com/montasim/b4joinacompany.git
cd b4joinacompany
pnpm install
cp .env.example .env.local
pnpm dev

Open http://localhost:3000.

The repository includes a bundled dataset in data/. During development the app prefers ../github-dataset-release/data when that sibling dataset project is present. Set DATASET_ROOT to use another release directory.

Environment variables

The defaults in .env.example are enough to inspect the public research experience with a local MongoDB instance. Configure the remaining integrations as needed:

Variable	Purpose
MONGODB_URI, MONGODB_DB	Authentication, saved checkpoints, generation quotas, and operational metadata
BETTER_AUTH_SECRET	Better Auth signing secret; use at least 32 random characters outside local development
BETTER_AUTH_URL, NEXT_PUBLIC_APP_URL	Authentication callback and public application URLs
GOOGLE_CLIENT_ID, GOOGLE_CLIENT_SECRET	Google sign-in and private saved workspaces
GEMINI_API_KEY, GEMINI_MODEL	Primary optional answer provider
GROQ_API_KEY, GROQ_MODEL	Optional fallback answer provider
AI_PROVIDER	Preferred provider: gemini or groq
AI_DAILY_LIMIT, AI_MONTHLY_LIMIT	Per-actor Ask limits when MongoDB is configured
DATASET_ROOT	Optional path to a dataset release containing data/

Production builds require the authentication URL, app URL, secret, and Google credentials. Ask still works without an AI key by using the deterministic fallback.

.env.example currently includes AI_PROVIDER_ORDER, but the application does not read it; provider selection is controlled by AI_PROVIDER. NODE_ENV is set by the runtime and enables the production configuration checks.

Deployment

The maintained deployment is hosted on Netlify, using the build command and runtime versions declared in netlify.toml. For another deployment, provide the production authentication origin, Google OAuth callback credentials, MongoDB, and any optional AI provider keys in the host's secret store. Keep BETTER_AUTH_URL and NEXT_PUBLIC_APP_URL on the same canonical HTTPS origin.

Development commands

pnpm dev        # start the development server
pnpm build      # create a production build
pnpm start      # run the production build
pnpm lint       # run ESLint
pnpm typecheck  # check TypeScript
pnpm test       # run the Vitest suite
pnpm check      # typecheck, lint, test, and build

API

The versioned API powers both the website and the companion extension.

Method	Route	Purpose
GET	/api/v1/health	Dataset and provider readiness
GET	/api/v1/companies?q=...	Search company identities and aliases
GET	/api/v1/companies/:slug	Company identity, destinations, and work-arrangement evidence
GET	/api/v1/companies/:slug/stories?q=...	Public source excerpts
GET	/api/v1/companies/:slug/hiring	Dated hiring signals
GET	/api/v1/companies/:slug/salary	Source-attributed, unverified role salary ranges
POST	/api/v1/ask	Provider-neutral cited answer
GET/POST	/api/v1/workspace/checkpoints	Authenticated private checkpoints
PATCH	/api/v1/workspace/checkpoints/:id	Optimistic checkpoint revision update
POST	/api/v1/corrections	Submit a correction for manual review
GET/POST	/api/v1/extension/*	Extension-ready company, story, job, and Ask responses

Public story responses expose short excerpts and original links, never the private raw body. API errors use:

{
  "error": {
    "code": "ERROR_CODE",
    "message": "Human-readable description",
    "requestId": "request identifier"
  }
}

Dataset updates

Dataset collection and publishing are intentionally separate from this app. After producing and validating a release in the sibling dataset repository, copy its release artifacts into this project and run the full check:

rsync -a ../github-dataset-release/data/ ./data/
pnpm check

The web application only reads those files. Corrections submitted through the product remain pending records until a maintainer reviews them for a future dataset release.

Documentation

• Product and implementation context
• Prototype notes
• Companion browser extension
• Public methodology
• Versioned API reference
• Dataset update workflow
• Evidence and source handling

Support

Questions, corrections, and reproducible bug reports can be opened through GitHub Issues. Do not post private workplace details, credentials, or vulnerability information in a public issue. The in-product correction form is the appropriate route for disputing displayed evidence.

Contributions should preserve source traceability, keep reported and official information separate, and include the relevant pnpm check result.

The repository does not yet contain dedicated CONTRIBUTING.md, SECURITY.md, CODE_OF_CONDUCT.md, or SUPPORT.md files. Use Issues for public bugs and methodology discussion, Pull Requests for reviewable changes, and the maintainer's profile for private security or personal-data reports.

Funding

Optional SupportKori contributions help fund dataset review, hosting, source verification, and maintenance. Evidence corrections, careful issue reports, and documentation contributions are equally valuable.

License

This repository does not currently include a license file. Copyright remains with the author, and no open-source license should be assumed.

Maintainer

Mohammad Montasim Al Mamun Shuvo