Insights Methodology

Data Sources

All five charts read from three production tables. No personally identifiable information ever leaves the database — every endpoint emits only roll-ups (counts, sums, averages, percentages). Student names, emails, phone numbers and resume contents are never exposed by this dashboard.

• jobs — recruiter postings (title, company, salary text, location, college scope, created_at).
• job_applications — the student funnel: status moves through applied → shortlisted → selected → rejected with timestamps on each transition.
• colleges — joined for the state-level placement chart via jobs.college_id.

Definitions

"Selected"

A row in job_applications whose status is 'selected' and whose selected_at timestamp falls inside the chart's recency window. This is the recruiter pressing "Select" — it is not a signed offer letter and it is not a joining record.

"Package / LPA"

The jobs.salary column is unstructured text (recruiters type values like "5-7 LPA", "₹12 lakhs", "500000 INR"). We parse each value into a single LPA number (taking the midpoint of ranges). Rows that don't parse are excluded from package averages but still counted in posting counts. Therefore the LPA figures are approximations, not contractual amounts.

"Skill demand"

A naive case-insensitive keyword scan of jobs.title + jobs.description against a fixed 12-entry vocabulary: Java, Python, React, SQL, Node.js, AWS, Spring, .NET, Go, JavaScript, TypeScript, Kubernetes. Each posting contributes at most once per skill. "Growth" is the percentage change in mentions between the last 6 months and the preceding 6 months.

"Placement %"

For each state we compute (selected ÷ applied) × 100, clamped to the 0-100 range. States with fewer than 5 applications are filtered out as low-signal. A state's percentage tells you how successful applications from colleges in that state were — it is not the share of all graduates that got placed.

Recency Window

• Top Hiring Companies (monthly) — last 6 calendar months of selected_at timestamps.
• Average Package by Bucket — all currently-active job postings.
• In-Demand Skills — last 6 months for the demand count, preceding 6 months for the growth comparison.
• Top Recruiters (quarterly) — last 4 quarters of selections.
• Placement % by State — all applications regardless of date (state roll-ups are slow to move; recency adds noise).

Privacy & Aggregation Floor

Endpoints emit only roll-ups (counts, averages, percentages). No row contains a name, email, phone, resume content or any other student identifier. The placement-percentage chart additionally enforces a minimum of 5 applications per state before that state is included, to avoid effectively-identifying small cohorts. Cached responses (when enabled by the platform team) are also aggregation-only.

Limitations & Disclaimer

1. These charts are not a substitute for verified placement records. Treat them as directional signals of recruiter activity on the platform.
2. Salary parsing is best-effort. Postings with ambiguous or out-of-band salary text are excluded from package averages — this can bias the bucket distribution upward if low-end recruiters under-specify pay.
3. Skill counts depend on recruiters mentioning the keyword in the posting text. Frameworks, libraries and tools outside our 12-entry vocabulary are invisible to the chart; absence of a skill does not mean low demand.
4. State-level percentages depend on colleges.state being populated. Colleges without a state recorded are dropped from that chart.
5. All comparisons are within the New Leaf platform only — they say nothing about hiring outside the platform.