|
Download docs/PITCH_OUTLINE.md from deeep-sec/campus-placement-engine: direct link, hf CLI and curl.
- Browser
- Download file 6.73 kB
-
https://huggingface.co/deeep-sec/campus-placement-engine/resolve/main/docs/PITCH_OUTLINE.md
- Command line
-
hf download hf://deeep-sec/campus-placement-engine/docs/PITCH_OUTLINE.md
-
curl -L -o PITCH_OUTLINE.md https://huggingface.co/deeep-sec/campus-placement-engine/resolve/main/docs/PITCH_OUTLINE.md
6.73 kB
Institutional Placement Engine: Pitch Presentation Outline
Competition: Campus Placement System
Problem Statement: Core Engine ("One student, three drives, one Tuesday")
Tagline: Deterministic Eligibility. 1-Second Clash-Free Scheduling. 100% Air-Gapped.
Slide 1: Title & The Tuesday Morning Nightmare
- Visual: Split graphic: on the left, chaotic sticky notes, competing spreadsheets, and frantic coordinators; on the right, the clean, unified Institutional Placement Engine.
- Talking Points:
- Collegiate placement departments orchestrate the most consequential weeks of their students' academic lives under acute operational pressure.
- 600 candidates, 12 concurrent companies, 8 physical rooms, and shifting corporate cut-offs.
- A student holding an offer at 10 AM is legally blocked by institutional policy for an afternoon drive.
- Room double-bookings lead to canceled corporate rounds; manual eligibility mistakes spark student grievances and accusations of coordinator bias.
Slide 2: Why Legacy Tools & Cloud Platforms Fail
- Visual: Comparison matrix: Legacy Spreadsheets vs Enterprise SaaS vs Core Engine Placement Engine.
- Talking Points:
- Spreadsheets: Silent formula errors, lack of audit trails, impossible concurrent conflict resolution.
- Commercial Cloud Portals: Require internet access (failing during campus network brownouts), leak student PII to commercial clouds, and lack deterministic constraint solvers.
- The Requirement: A single-laptop, 100% offline engine that guarantees mathematical rigor, complete data sovereignty, and zero-clash operations.
Slide 3: The Architectural Blueprint: Pure Domain Core
- Visual: C4 Container Diagram illustrating Pure Domain Core decoupled from Storage via Repository Protocols.
- Talking Points:
- Pure Functional Core: Eligibility rules, offer policies, scheduling constraints, and what-if simulations run as pure deterministic Python functions with zero database I/O.
- Single-Port LAN Serving: A single command boots both the FastAPI backend and bundled React SPA on port 8000, broadcasting a local URL for student smartphones on the campus Wi-Fi.
- Relational Integrity: SQLite in Write-Ahead Logging (WAL) mode with strict foreign key constraints and optimistic concurrency control (
versionstamps).
Slide 4: Key Innovation 1 — Typed Rule AST & Empathic Explainability
- Visual: Screenshot of AST Node Evaluation and the Two-Tier Explanation Engine.
- Talking Points:
- 13 Rule Primitives: CGPA, 10th/12th/diploma percentages, branch aliases, active/historical backlogs, gap years, and boolean groups.
- Demographic Isolation: Protected demographic attributes (
gender,caste,religion,category) are barred from eligibility logic at the type-system level. - Two-Tier Explanations:
- Candidate: Respectful, single-line explanation: "Not eligible: your CGPA is 6.8 and this company requires 7.0 or above."
- Officer: Complete cryptographic trace citing rule ID, database source, actual vs cut-off values, and SHA-256 data snapshot hash.
Slide 5: Key Innovation 2 — Deterministic Document Extraction
- Visual: Side-by-side view of raw messy PDF circular and parsed AST fields with confidence badges.
- Talking Points:
- University TPOs receive circulars as messy emails, PDFs, and Word documents.
- 100% Deterministic & Offline: Zero LLM hallucinations, zero cloud APIs, zero token costs.
- Benchmark Validation: Successfully parsed 25 out of 25 benchmark announcements (100.0%) without manual re-typing, averaging 14.6 milliseconds per document.
Slide 6: Key Innovation 3 — 1-Second Constraint Scheduling (OR-Tools CP-SAT)
- Visual: Interactive Rooms $\times$ Time Schedule Grid with "0 Clashes Verified" Badge.
- Talking Points:
- Modeled as a formal Constraint Satisfaction Problem (CSP) in Google OR-Tools CP-SAT.
- Hard Constraints:
NoOverlapon physical rooms,NoOverlapon company panels,NoOverlapon candidates across concurrent drives, plus 15-minute inter-block walking buffers. - World-Class Performance:
- 600 students across 12 drives and 8 rooms solved in 1.05 seconds (Target: < 60s).
- 1,800 students across 40 drives and 20 rooms solved in 5.04 seconds.
Slide 7: Key Innovation 4 — Interactive What-If Simulation & Dynamic Headlines
- Visual: What-If panel showing dynamic headline and schedule displacement preview.
- Talking Points:
- Placement officers can test prospective changes in 18 milliseconds without modifying production state.
- Data-Driven Dynamic Headlines: Headlines like "Lowering cut-off from 7.0 to 6.5 adds 38 students; 6 clash with Northwind test at 10:00" are generated directly from data deltas.
- Publish Workflow: Monotonic versioning with immutable historical snapshots and instant one-click rollback.
Slide 8: Institutional Security & Air-Gap Compliance
- Visual: STRIDE Threat Model matrix and Automated Socket Interception Verification.
- Talking Points:
- Air-Gap Verification: Automated socket-level tests intercept OS network calls to mathematically prove zero outbound connections.
- Cryptographic Security: Argon2id password hashing, AES-256 Fernet field-level encryption, PBKDF2-encrypted
.encdatabase exports, and one-click cryptographic institutional wipe. - Candidate Privacy: Candidates authenticate via roll number + temporary code, accessing strictly their own schedule pass with zero access to peer rankings.
Slide 9: Measured Benchmark Scorecard
- Visual: Summary Scorecard comparing Core Engine targets against measured numbers:
- Batch Eligibility: 1,800 students in 0.024s (Target: < 1.0s).
- Scheduler: 600 students / 12 drives / 8 rooms in 1.05s (Target: < 60s).
- Extraction: 25 / 25 announcements (100%) (Target: $\ge$ 20/25).
- Air-Gap Connections: 0 Outbound Calls (Target: 0).
- Test Suite: 149 / 149 Pytest passing, 0 Ruff errors, 0 Mypy errors.
Slide 10: Conclusion & Institutional Vision
- Visual: Placement engine running live on a single laptop serving mobile browsers across an air-gapped auditorium.
- Closing Statement:
"The Institutional Placement Engine delivers institutional fairness, mathematical precision, and operational resilience. It transforms the chaotic Tuesday morning placement crisis into a calm, transparent, and completely clash-free triumph."