PRAMAAN-X

AI-Powered Regulatory Decision-Support Framework for AICTE Approval Document Scrutiny

Quick Links: Live Demo • Presentation Pitch Deck • Presentation Outline • Full Evaluation Report • 154-Element Audit Report • Deployment Guide • OCR Setup Guide

🌐 Live Production Demo

The PRAMAAN-X platform is live and operational on Render across three decoupled tiers:

Operational Tier	Service Component	Live Endpoint	Health & Telemetry	Technology Stack
Frontend	Scrutiny Console	pramaan-frontend.onrender.com	`Active`	React 19 + Vite + Tailwind CSS
Backend API	Core Regulatory Gateway	pramaan-backend-y3bl.onrender.com	`/health`	FastAPI + PostgreSQL + NetworkX
Mock Authority	Statutory Issuer Service	pramaan-mock-issuer.onrender.com	`/health`	FastAPI + 510-Record Registry

🔑 Demo Login Credentials: To evaluate immediately, log in as Scrutiny Officer using scrutiny.officer@aicte-india.org / AicteOfficer@2025 (additional roles detailed in Demo Accounts).

⚡ In 30 Seconds

The Problem: Over 10,000 technical institutions submit multi-document regulatory dossiers annually for AICTE Extension of Approval (EoA). Manual committee review faces an assumed baseline of 180 minutes per dossier, creating backlogs while struggling to catch cross-institution fraud (e.g. "ghost faculty" claimed concurrently in multiple colleges or facilities sharing identical photos across states).
Our Solution: PRAMAAN-X is an advisory decision-support platform that audits compliance dossiers in 2.03 seconds, highlights discrepancies with exact PDF bounding boxes, cross-references digital twins, and never automatically rejects an institution. Final statutory authority stays strictly with human scrutiny officers.
What Makes It Different:
1. Regulation-as-Code: AICTE Approval Process Handbook (APH) clauses codified as declarative YAML with cohort "What-If" impact dry-runs.
2. Impossibility Engine (NetworkX): Multi-institutional knowledge graph uncovering dual-employment collisions, land overlap, and chronological impossibilities.
3. Photo Reuse Radar & Digital Forensics: 64-bit perceptual hashing (pHash), Error Level Analysis (ELA), and copy-move forgery detection.
4. Tamper-Evident Verification Ledger: Append-only SHA-256 Merkle chain with asymmetric Ed25519 digital signatures and database triggers.
5. Pluggable OCR Layer: Transparent multi-engine text extraction (lossless PyMuPDF text layer for digital vector PDFs; RapidOCR with OpenCV deskew/CLAHE for scans and images).

📸 Screenshots Gallery & Interactive UI Tour

1. Primary Operational Scrutiny Views

Screen & Focus	Screen & Focus
Risk-Ranked Scrutiny Queue: Real-time triage queue, KPI cards, anomaly severity distribution, and instant filter by deficiency type.	Evidence Lens (Split-Viewer): Synchronized PDF canvas with bounding-box OCR highlights, regulatory clause citations, and officer action drawer.
Pre-Flight Readiness Scorecard: Self-service institutional portal with deficit radar, mandatory checklist tracking, and corrective hints.	National Photo Reuse Radar: Interactive cluster graph discovering identical campus facilities and lab equipment claimed across separate colleges.

2. Deep Regulatory & Forensic Intelligence Views

Screen & Focus	Screen & Focus
Impossibility Engine (vis-network): Relational twin graph flagging concurrent faculty dual-employment and spatial land plot overlap.	Document Ingestion & Forensics: Live PyMuPDF text parser, Error Level Analysis (ELA) heatmap scanner, and certificate integrity check.
Regulation-as-Code & What-If Simulator: Declarative YAML handbook rule editor with hot reload and cohort-wide failure simulation.	Geospatial Inspection Planner: Risk-prioritized regional inspection itinerary clustering high-deficit colleges for surprise physical visits.
Manual Review Queue: Dedicated workflow for ambiguous or degraded certificates requiring officer secondary determination.	National Institutional Design System: Accessible UI meeting WCAG 2.1 AA contrast standards with seamless Dark Mode support.

3. Live Pluggable OCR Scrutiny Evidence

Live Digital PDF (Text Layer)	Live Scanned PDF (RapidOCR)
PyMuPDF Text Layer: Vector text extraction displaying `Engine: PyMuPDF text layer` (mean confidence 99.0%).	RapidOCR on Degraded Scan: Raster PDF rendered at 250 DPI with OpenCV deskew displaying `Engine: RapidOCR` (mean confidence 96.7%).
Live Standalone Image (PNG / JPG)	Live Institutional Pre-Flight Portal
RapidOCR on Image: Pure PNG certificate extracted via ONNX Runtime displaying `Engine: RapidOCR` (mean confidence 97.3%).	Pre-Flight Institutional Portal: Live backend connectivity (`Live Backend` green pill) showing active file ingestion.

4. Cryptographic Tamper-Detection Demonstration

PRAMAAN-X features an active mathematical verification ledger. The interactive tamper demo proves immediate severed-link detection if any record is mutated:

Step 1: Pristine Hash Chain Verified	Step 2: Adversarial Mutation Trigger
All Blocks Authentic: SHA-256 Merkle chain and Ed25519 digital signatures verify authentic across all blocks.	Tamper Simulation Modal: Evaluator injects an adversarial payload modification into Block #2.
Step 3: Cryptographic Link Breach Pinpointed	Step 4: Pristine Ledger Restored & Re-Verified
Link Severed: Modifying Block #2 payload instantly invalidates downstream Block #3 parent hash in red.	Restoration Verified: Re-verifying pristine ledger plays green validation animation and restores integrity.

5. Mobile Responsive Experience (Phones & Tablets)

PRAMAAN-X provides full responsive access across mobile viewports (360px–430px) and tablets (768px), with off-canvas navigation drawers, touch targets $\ge 44\times 44\text{px}$, responsive card transformations for triage queues, and zero horizontal scroll:

Mobile Dashboard & Triage Cards	Off-Canvas Navigation Drawer
Mobile Scrutiny Dashboard: Compact 2-column KPI grid, mobile risk pills, and adaptive application cards with full touch targets.	Mobile Slide-Out Drawer: Instant category navigation, active tab indicator, role badge, and backdrop-blur overlay.
Mobile Evidence Lens	Mobile Photo Radar & Clustering
Mobile Evidence Lens: Stacked split-view with synchronized PDF viewport, clause citations, and one-tap officer actions.	Mobile Photo Radar: Responsive reuse cluster analysis with comparative thumbnail cards and similarity metrics.

⚙️ How It Works: The 4-Step Pipeline

  1. INGEST & EXTRACT           2. MULTI-ENGINE SCRUTINY        3. AUTHORITY LOOKUP         4. RECEIPT & DECISION
┌──────────────────────┐      ┌───────────────────────────┐   ┌───────────────────────┐   ┌──────────────────────────┐
│ Compliance Dossier   │ ───► │ • Regulation-as-Code      │ ─►│ Mock Statutory Issuer │──►│ • Ed25519 Signed Receipt │
│ • 5 Statutory PDFs   │      │ • Impossibility Engine    │   │ • State Fire Services │   │ • Bounding Box Evidence  │
│ • Pluggable OCR Layer│      │ • Photo Radar & Forensics │   │ • Land Revenue        │   │ • Officer Review & Action│
└──────────────────────┘      └───────────────────────────┘   └───────────────────────┘   └──────────────────────────┘

Document Ingestion & Field Extraction (backend/app/services/ocr/, backend/app/services/):
Ingests statutory compliance PDFs and images (Fire NOC, Land Deed, Affiliation, Building Plan, Affidavit). The pluggable OCR layer routes digital PDFs through PyMuPDF text layer (lossless vector extraction) and scans/images through RapidOCR (rapidocr-onnxruntime with OpenCV deskew/CLAHE). Deterministic regex extractors parse sanctioned intake, carpet areas, survey numbers, and validity dates with bounding boxes ($F_1: 0.9013$ on held-out Seed 7; $F_1: 0.969$ on scanned benchmark).
Multi-Engine Scrutiny & Cross-Institution Analysis (backend/app/rules/, backend/app/services/):
- Evaluates 12 declarative APH handbook rules (FSR 1:20, carpet area $\ge 10.5\text{ m}^2/\text{student}$, cadre ratios).
- Projects colleges into an in-memory NetworkX digital twin to detect faculty sharing (Ghost Faculty) and pre-dated certificates.
- Computes 64-bit image perceptual hashes (pHash) to discover identical campus facilities claimed across institutions.
Statutory Authority Cross-Verification (mock_issuer/, backend/app/services/connectors/):
Dispatches record lookups against state authorities (Fire Services, Affiliating University, Land Revenue/Bhoomi, NBA). Performs QR payload parsing (pyzbar) and PDF digital signature byte-range structural inspection.
Receipt Minting & Human Officer Determination (backend/app/services/, backend/app/api/):
Mints an asymmetric RFC 8032 Ed25519-signed verification receipt appended to a SHA-256 Merkle chain. Surfaces findings on the Evidence Lens for officer determination; mandatory justification comments are enforced on any adverse action.

🏗️ System Architecture

Logical Service Blueprint & Dataflow	Live Multi-Tier Architecture & Node Inspector
Analytical Core Topology: FastAPI gateway orchestrating the 6 statutory engines, SQLite/PostgreSQL persistence, and local mock statutory issuer.	Interactive System Inspector: Live tier inspection with performance SLAs, service interfaces, technology stacks, and statutory safeguards.

flowchart TB
    subgraph CLIENT["Presentation Layer"]
        UI["React 19.3.1 + Tailwind CSS + Vite<br/>11 Scrutiny Views • Evidence Lens • vis-network"]
    end

    subgraph API["API Gateway & Security"]
        GATEWAY["FastAPI 0.115.6 Gateway<br/>85 Endpoints • JWT Bearer • RBAC • Tenant Isolation"]
    end

    subgraph ENGINES["6 Analytical & Forensic Engines"]
        direction TB
        E1["Checklist Engine<br/>Application Category Schemas"]
        E2["Pluggable OCR & Extraction Engine<br/>PyMuPDF Text Layer + RapidOCR (eng)"]
        E3["Regulation-as-Code Engine<br/>12 APH Rules & Cohort What-If Sim"]
        E4["NetworkX Impossibility Engine<br/>Multi-Campus Twin & Ghost Faculty"]
        E5["Forensics & Photo Radar<br/>ELA, Copy-Move, 64-bit pHash (Hamming <= 10)"]
        E6["Verification Ledger Engine<br/>Ed25519 Receipts & SHA-256 Merkle Chain"]
    end

    subgraph ISSUER["Statutory Verification Gateway"]
        MOCK["Mock Statutory Issuer (Port 8001)<br/>510 Authority Records (Fire, University, Revenue, NBA)"]
    end

    subgraph DATA["Data & Persistence Layer"]
        DB[("SQLite Local / PostgreSQL Compose<br/>Append-Only DB Triggers")]
        RULES["Declarative YAML Packs<br/>aicte_aph_2025.yaml"]
        GRAPH["In-Memory Twin Graph<br/>NetworkX MultiDiGraph"]
    end

    UI -->|HTTP / JSON REST| GATEWAY
    GATEWAY --> ENGINES
    E1 & E2 & E3 & E4 & E5 & E6 <--> DATA
    E1 & E3 & E6 <-->|Local Authority Verification| MOCK

Evidence Lens Regulatory Traceability Flow

sequenceDiagram
    autonumber
    actor Officer as Scrutiny Officer
    participant UI as Evidence Lens UI
    participant Core as Scrutiny Pipeline
    participant Rule as Regulation-as-Code
    participant Issuer as Mock Statutory Issuer
    participant Ledger as Receipt Ledger

    Officer->>UI: Selects Institutional Dossier
    UI->>Core: Fetch Extracted Claims & Bounding Boxes
    Core->>Rule: Evaluate APH Handbook Rules (FSR, Density, NOC Age)
    Core->>Issuer: Query Authority Record (/verify/authority/id)
    Issuer-->>Core: Registry Match (Verified / Mismatch / Expired)
    Core->>Ledger: Mint Ed25519 Signed Verification Receipt
    Ledger-->>Core: Block Appended to SHA-256 Merkle Chain
    Core-->>UI: Render Claim ➔ Evidence ➔ Rule Citation ➔ Source Status
    Officer->>UI: Select Action (Approve / Reject / Seek Clarification) + Mandatory Comment
    UI->>Ledger: Commit Officer Determination (No Auto-Rejection)

📊 Benchmark Results & Performance Charts

All quantitative figures below are extracted directly from evidence_pack/metrics.json and docs/evaluation_report.md.

Metric Dimension	Measured Value	Benchmark Context
Held-Out Test Set	100 institutions / 500 documents	Seed 7 held-out test split (calibrated on Seed 42)
Planted Defect Cases	240 positive cases	Distributed across 20 distinct statutory violation categories
Clean Control Cases	57 clean cases	55 True Negatives, 2 False Positives (conservative flags on noisy scans)
True Positives (TP) / False Negatives (FN)	TP: 240 / FN: 0	100.0% Recall across all 20 statutory defect categories
False Positives (FP) / True Negatives (TN)	FP: 2 / TN: 55	0.0351 False-Positive Rate (FPR: 3.51%)
Defect Detection Precision	0.9917 (99.17%)	$240 / (240 + 2)$
Defect Detection F1-Score	0.9959	Harmonic mean of precision (0.9917) and recall (1.0000)
Held-Out Field Extraction Macro F1	0.9013	Macro average across 6 statutory certificate fields on Seed 7
Held-Out Field Extraction Micro F1	0.8596	Global micro $F_1$ ($TP: 199, FP: 38, FN: 27$)
Scanned/OCR Path Macro F1 (20-Doc Benchmark)	0.969 (Digital: 0.982)	RapidOCR evaluated on pure raster degraded scans vs PyMuPDF vector text
Automated Test Suite Status	182 / 182 passing (100%)	15 test suites across `backend/tests` and `mock_issuer` (+ 7 in `backend/test_api.py` = 189 tests total)
Automated Pipeline Processing Time	2.03 seconds	Measured mean runtime across 6 evaluation stages
Alert Dispatch Latency	Mean: 3.21 ms / Median: 3.20 ms	In-app WebSocket notification and event bus latency

⚠️ Important Disclosure: Synthetic benchmark (calibrated on Seed 42, evaluated on locked held-out Seed 7), not field accuracy on live production AICTE filings. The 180-minute manual scrutiny baseline is an assumed operational comparison, not an empirically timed study.

Slide-Ready Evaluation Visualizations

Extraction Accuracy per Field	Defect Accuracy Across 20 Categories
Macro $F_1$: 0.9013 across certificate numbers, dates, acreage & surveys.	100.0% Recall across all 20 planted defect categories (240/240).

2.03s vs 180m Manual Baseline	Photo Radar & Anomaly Distribution
Automated pipeline turnaround of 2.03s per full dossier.	pHash precision: 0.8824, recall: 1.0000; anomaly precision: 0.9917.

Click to view Complete 20-Category Defect Coverage Breakdown (from evidence_pack/defect_coverage.csv)

#	Defect Identifier	Planted	Detected	Accuracy	Detection Module & Mechanism
1	`expired_document`	19	19	100.0%	Rule Engine: `RULE_FIRE_NOC_VALIDITY` (Cl 6.3.2) cutoff date check
2	`missing_signature_field`	22	22	100.0%	Forensics / Connector: `DigitalSignatureConnector` notary check
3	`wrong_institution_name`	15	15	100.0%	Rule Engine: `RULE_NAME_MATCH_AFFIDAVIT_NOC` (token similarity < 0.85)
4	`shared_document_reference_number`	3	3	100.0%	Impossibility Engine: Multi-dossier reference ID collision
5	`faculty_duplicate_across_institutions`	21	21	100.0%	Impossibility Engine: Ghost Faculty cross-college PAN/Aadhaar collision
6	`area_too_small_for_intake`	17	17	100.0%	Rule Engine: `RULE_BUILT_UP_AREA_PER_STUDENT` (< 10.5 sqm/student)
7	`date_before_issuer_existed`	21	21	100.0%	Impossibility Engine: Temporal causal check (issue date predates issuer)
8	`inverted_validity_interval`	3	3	100.0%	Forensics: Temporal validator (issue date postdates expiry date)
9	`reused_photo`	42	42	100.0%	Photo Reuse Radar: 64-bit pHash Hamming distance ($\le 10$)
10	`missing_exif_metadata`	21	21	100.0%	Forensics: EXIF inspector (camera make, timestamp, GPS tags missing)
11	`pdf_metadata_producer_anomaly`	3	3	100.0%	Forensics: PDF structural scan (Ghostscript/iText trailer tampering)
12	`font_inconsistency_in_region`	3	3	100.0%	Forensics: Typography uniformity scan (font family and size anomaly)
13	`error_level_analysis_spliced_patch`	3	3	100.0%	Forensics: Error Level Analysis compression artifact variance (> 0.65)
14	`copymove_cloned_seal`	3	3	100.0%	Forensics: ORB keypoint duplicate stamp and seal matcher
15	`qr_ocr_reference_mismatch`	3	3	100.0%	Forensics: `QrDecodeConnector` (pyzbar) vs OCR extracted certificate ID
16	`sun_shadow_angle_discrepancy`	3	3	100.0%	Proof-of-Reality: `sun_shadow_service` (pvlib solar ephemeris vs shadow)
17	`built_up_area_footprint_discrepancy`	3	3	100.0%	Proof-of-Reality: `built_up_area_service` (satellite polygon vs claim)
18	`lab_equipment_inconsistent`	2	2	100.0%	Impossibility Engine: Equipment inventory serial number collision
19	`mismatched_reference_number`	20	20	100.0%	Rule Engine: `RULE_SURVEY_NUMBER_CONSISTENCY` (Cl 7.2.1) cross-check
20	`edited_amount`	13	13	100.0%	Rule Engine / Forensics: Leasehold tenure (< 30 years) and area statement

🛡️ Statutory Safeguards & Verifying Tests

PRAMAAN-X strictly upholds statutory administrative fairness principles. Automated checks never issue punitive verdicts.

Safeguard Policy	Code Implementation	Verifying Test Suite & File Path
No Automatic Rejection Adverse determination requires officer action with mandatory comment	`models.py:140-155` Empty comment rejected with HTTP 400	`test_no_automatic_rejection_rule` `backend/tests/test_rbac_and_endpoints.py:182` `test_adverse_status_impossible_without_officer_action` `backend/tests/test_rectification_and_features.py:263`
RBAC & Tenant Isolation 5 roles; colleges cannot see other colleges' dossiers	`security.py`, `auth.py` Institutional query scoping	`test_login_all_5_roles` `backend/tests/test_rbac_and_endpoints.py:70` `test_institution_user_cannot_view_other_institution_data` `backend/tests/test_rbac_and_endpoints.py:112`
PII Data Masking Masks national PAN / Aadhaar for officers and auditors	`field_extraction.py:32` `mask_pan` and `mask_aadhaar`	`test_pii_masking_for_officers_and_auditors` `backend/tests/test_rbac_and_endpoints.py:284`
Sensitive Access Logging Audit log entry recorded on all sensitive reads	`models.py` `AuditLog` table with timestamp	`test_access_logging_for_sensitive_reads` `backend/tests/test_rbac_and_endpoints.py:320`
Advisory Language Mandate Strict non-punitive wording ("Indicator for review")	Hardcoded policy in pipeline services Prohibits verdict terms	`test_mandatory_advisory_caveat` `backend/tests/test_sun_shadow_plausibility.py:199`
Tamper-Evident Ledger Security Append-only ledger with cryptographic hash chaining	`verification_receipts.py` Ed25519 signatures, DB triggers	`test_tamper_detection_content_modification` `backend/tests/test_verification_receipts.py:90` `test_database_trigger_blocks_direct_sql_mutation` `backend/tests/test_verification_receipts.py:130`

🔍 What is Real vs. Simulated

We believe in complete technical transparency with Hackathon evaluators.

Component / Capability	Implementation Reality	Current Repository State
Text Stream Extraction	REAL	PyMuPDF (`fitz`) direct stream parsing + deterministic regex boundary normalizers.
Regulation-as-Code	REAL	Declarative YAML parser, JSON Schema pack validator, and cohort What-If simulation.
Impossibility Engine	REAL	In-memory NetworkX graph analyzing dual employment, spatial overlap, and chronology.
Photo Reuse Radar	REAL (pHash) / SIMULATED (Vector)	Perceptual 64-bit `imagehash` (pHash/dHash) is real. OpenCLIP 512-dim cosine similarity is simulated via projection matrices.
Digital Forensics	REAL	Error Level Analysis (ELA), copy-move keypoint matching, and PDF metadata inspection.
QR Code Extraction	REAL	Native 2D barcode rasterization and decoding via `pyzbar` and OpenCV.
Digital Signatures	HYBRID	PDF `/ByteRange` structural integrity inspection is real. X.509 PKI trust anchor chain validation is a stub.
Verification Receipts	REAL	Asymmetric Ed25519 digital signatures (RFC 8032) and SHA-256 Merkle chain with SQLite/Postgres DB triggers.
Statutory Authority Gateways	SIMULATED	Local Mock Statutory Issuer (`127.0.0.1:8001`) with 510 pre-seeded JSON records. Live API Setu and DigiLocker are not integrated.
OCR for Scans & Images	REAL (PyMuPDF + RapidOCR English-only)	Installed and verified on this machine: PyMuPDF text layer (digital vector PDFs) + RapidOCR (`rapidocr-onnxruntime==1.4.4`, English only) with OpenCV preprocessing. Tesseract binary is NOT installed on this host (optional fallback when `tesseract.exe` is found).
APH 2025-26 Rule Parameters	PLACEHOLDERS	All 12 rule parameters in `aicte_aph_2025.yaml` are engineering placeholders awaiting gazetted sign-off.
Docker Compose	VALIDATED / NOT VERIFIED	`docker-compose.yml` validates syntactically; runtime container build was not verified locally due to inactive daemon.
Manual Scrutiny Baseline	ASSUMED	The 180.0 min manual review baseline is an assumed benchmark figure, not an empirical time study.

🔍 Pluggable OCR Architecture for Scans & Images

PRAMAAN-X features a unified pluggable OCR layer (backend/app/services/ocr/):

PyMuPDF Text Layer: Used when a PDF page contains embedded vector text (lossless character stream, 0.990 confidence).
RapidOCR (rapidocr-onnxruntime==1.4.4): Pip-installable deep learning OCR engine (installed & verified on this host, English only), requiring zero external native C++ binary installers. Includes OpenCV preprocessing (deskew via minAreaRect, CLAHE contrast enhancement).
Tesseract 5.x: Autodetected via TESSERACT_PATH, Windows default (C:\Program Files\Tesseract-OCR\tesseract.exe), or system PATH. Note: Tesseract binary is NOT installed on this Windows host; RapidOCR provides the primary image extraction.
Graceful Fail-Safe: If no engine is available for an image, returns status: "ocr_unavailable" with an explicit error banner rather than silent empty text. Low-confidence extractions automatically route to the Manual Review Queue.

Operational Limitations & Caveats:

Handwriting: Handwritten signatures, marginalia, and endorsement notes are not recognized by the printed OCR models and are flagged for Scrutiny Officer manual review.
Extreme Scan Degradation: Severely crumpled, torn, or low-resolution scans (< 100 DPI) produce lower confidence and are safely diverted to manual review.
Regional Languages: Hindi (hin), Kannada (kan), and Tamil (tam) require their respective .traineddata models installed in Tesseract's tessdata directory. RapidOCR currently operates on default English recognition weights.
Synthetic Benchmark Caveat: Standard benchmark evaluations are conducted on synthetically generated certificates calibrated to AICTE APH 2025-26 statutory clauses.

🚀 Quick Start Guide

Prerequisites

Python 3.12+
Node.js 18+ and npm

1-Click Launch (Recommended)

Run the automated launcher from repository root:

# Windows:
run_all.bat

run_all.bat automatically launches the Mock Issuer (8001), FastAPI Backend (8000), and Vite Frontend (5173), then opens the browser.

Manual Step-by-Step Launch

If launching services individually across terminal windows:

# Terminal 1: Mock Statutory Issuer Gateway (Port 8001)
python -m uvicorn mock_issuer.app.main:app --host 127.0.0.1 --port 8001

# Terminal 2: FastAPI Backend API (Port 8000)
python -m uvicorn backend.app.main:app --reload-dir backend --host 127.0.0.1 --port 8000

# Terminal 3: Vite React Frontend (Port 5173)
cd frontend && npm run dev -- --host 127.0.0.1 --port 5173

Then open: http://localhost:5173

💡 Troubleshooting: If you see ECONNREFUSED 127.0.0.1:8000, the FastAPI backend process has stopped or has not completed initialization. Confirm port 8000 is open.

👥 Demo Accounts (Synthetic Testing Credentials)

The system provides 5 mock role presets populated with synthetic data:

Role Title	Email Credential	Password	Primary Accessible Scope
Scrutiny Officer	`scrutiny.officer@aicte-india.org`	`AicteOfficer@2025`	Desk scrutiny queue, Evidence Lens, bounding-box review
Regional Officer	`regional.officer@aicte-india.org`	`AicteRegional@2025`	Regional escalations, approvals, What-If cohort simulations
Central Admin	`admin@aicte-india.org`	`AicteAdmin@2025`	Rule catalog editor, handbook parameter hot-reload, users
Independent Auditor	`auditor@aicte-india.org`	`AicteAuditor@2025`	Read-only tamper-evident audit ledger and cryptographic receipts
Institution User	`registrar@apextech.edu.in`	`ApexPass@2025`	Apex Institute of Technology dossier, Pre-Flight Scorecard

Note: All credentials above are mock accounts intended for evaluation and demonstration.

💻 Tech Stack & Repository Structure

Core Technologies

Frontend: React 19.3.1, Vite 8.3.4, Tailwind CSS 3.4.19, vis-network 10.1.3, Lucide Icons
Backend API: FastAPI 0.115.6, Uvicorn 0.34.0, Pydantic 2.10.4, SQLAlchemy 2.0.36
Analytical & Graph: NetworkX 3.4.2, PyMuPDF 1.25.1, RapidOCR 1.4.4, ONNX Runtime 1.31.1, imagehash 4.3.1, OpenCV, Shapely, pvlib
Security & Integrity: Asymmetric Ed25519 (cryptography), SHA-256 Merkle chain, SQLite DB triggers
Python Environment: CPython 3.12.8 (64-bit)

Repository Directory Tree

SIH_AKXY/
├── backend/
│   ├── app/
│   │   ├── api/             # 85 REST API endpoints across 10 routers
│   │   ├── core/            # JWT authentication, RBAC policy, DB session
│   │   ├── models/          # SQLAlchemy schemas (Dossiers, Receipts, AuditLog)
│   │   ├── rules/           # AICTE APH 2025-26 YAML rule packs & JSON schemas
│   │   └── services/        # 6 core engines (Extraction, OCR, Forensics, Radar, Twin)
│   ├── tests/               # 14 automated test suites (173 tests)
│   └── test_api.py          # Integration endpoint test suite (7 tests)
├── mock_issuer/             # Mock Statutory Issuer service on port 8001 (9 tests)
├── frontend/
│   ├── src/
│   │   ├── components/      # EvidenceLens, PhotoRadar, Navbar, BoundingBoxViewer
│   │   ├── pages/           # 11 scrutiny screens (Dashboard, Preflight, Ledger)
│   │   └── services/        # API client and cold-start retry connectors
│   └── package.json
├── synthetic_generator/     # Synthetic dossier & photo defect generation tools
├── generated_data/          # Ground truth CSVs, catalogs, and test PDFs
├── evidence_pack/           # Ground truth metrics.json, defect_coverage.csv, reports
├── docs/                    # Architecture diagrams, screenshots, reports, presentations
│   ├── presentation/        # Slide deck (.pptx and .pdf)
│   ├── DEPLOY.md            # Multi-tier deployment & operations guide
│   └── AUDIT_REPORT.md      # 154-element end-to-end verification audit report
├── tools/                   # Smoke verification & deployment testing scripts
├── scripts/                 # Maintenance and demo database reset scripts
├── docker-compose.yml       # Production multi-tier container composition
├── render.yaml              # Infrastructure-as-Code blueprint for Render
├── run_all.bat              # 1-click Windows full-stack launcher
└── README.md

Running the Test Suite

To execute the automated regression test suite:

# Run all tests across backend and mock issuer:
$env:PYTHONPATH="."; pytest backend/tests mock_issuer/test_mock_issuer.py -q

Expected Result: 182 passed in backend/tests and mock_issuer (+ 7 in backend/test_api.py = 189 tests passed total).

🗺️ Phased Implementation Roadmap

Phase 1: Regulatory Calibration: Formal Gazette of India sign-off on the 12 Approval Process Handbook (APH) rule parameters and cutoff tolerances.
Phase 2: Authorized Scrutiny Pilot: Controlled pilot deployment on anonymized historical dossiers across select AICTE Regional Offices.
Phase 3: Production Authority Binding: Integration with live API Setu and DigiLocker production gateways; replacement of mock issuer.
Phase 4: Multilingual OCR Deployment: Packaging regional language traineddata (Hindi, Kannada, Tamil) with TrOCR/Tesseract models.
Phase 5: National Rollout: Phased migration across all regional scrutiny cells with HSM-backed Ed25519 signing keys.

🏆 Team AnomalyX & Statutory Disclaimer

Hackathon: Smart India Hackathon (SIH) 2026
Problem Statement: SIH26261 — AI-Powered Regulatory Scrutiny & Compliance Automation
Organization: All India Council for Technical Education (AICTE), Ministry of Education, Govt. of India
Team: AnomalyX

⚖️ Statutory Notice & Disclaimer: PRAMAAN-X is an academic research prototype developed for Smart India Hackathon 2026. All evaluation data and institutional filings are synthetic. PRAMAAN-X operates strictly as an advisory decision-support system and does not issue final statutory accreditation determinations.

Navigation Menu

Repository navigation

About

Repository files navigation

PRAMAAN-X

🌐 Live Production Demo

⚡ In 30 Seconds

📸 Screenshots Gallery & Interactive UI Tour

1. Primary Operational Scrutiny Views

2. Deep Regulatory & Forensic Intelligence Views

3. Live Pluggable OCR Scrutiny Evidence

4. Cryptographic Tamper-Detection Demonstration

5. Mobile Responsive Experience (Phones & Tablets)

⚙️ How It Works: The 4-Step Pipeline

🏗️ System Architecture

Evidence Lens Regulatory Traceability Flow

📊 Benchmark Results & Performance Charts

Slide-Ready Evaluation Visualizations

🛡️ Statutory Safeguards & Verifying Tests

🔍 What is Real vs. Simulated

🔍 Pluggable OCR Architecture for Scans & Images

🚀 Quick Start Guide

Prerequisites

1-Click Launch (Recommended)

Manual Step-by-Step Launch

👥 Demo Accounts (Synthetic Testing Credentials)

💻 Tech Stack & Repository Structure

Core Technologies

Repository Directory Tree

Running the Test Suite

🗺️ Phased Implementation Roadmap

🏆 Team AnomalyX & Statutory Disclaimer

About

Releases

Packages

Contributors

Languages

Name		Name	Last commit message	Last commit date
audit_fixtures		audit_fixtures
backend		backend
docs		docs
evidence_pack		evidence_pack
frontend		frontend
generated_data		generated_data
mock_issuer		mock_issuer
scripts		scripts
synthetic_generator		synthetic_generator
tools		tools
uploads		uploads
.env.example		.env.example
.gitignore		.gitignore
Makefile		Makefile
README.md		README.md
docker-compose.yml		docker-compose.yml
make.bat		make.bat
metrics.json		metrics.json
render.yaml		render.yaml
requirements-ml.txt		requirements-ml.txt
run_all.bat		run_all.bat
run_backend.bat		run_backend.bat
run_frontend.bat		run_frontend.bat
run_mock_issuer.bat		run_mock_issuer.bat
vercel.json		vercel.json

Latest commit

History

Folders and files

Repository files navigation

PRAMAAN-X

🌐 Live Production Demo

⚡ In 30 Seconds

📸 Screenshots Gallery & Interactive UI Tour

1. Primary Operational Scrutiny Views

2. Deep Regulatory & Forensic Intelligence Views

3. Live Pluggable OCR Scrutiny Evidence

4. Cryptographic Tamper-Detection Demonstration

5. Mobile Responsive Experience (Phones & Tablets)

⚙️ How It Works: The 4-Step Pipeline

🏗️ System Architecture

Evidence Lens Regulatory Traceability Flow

📊 Benchmark Results & Performance Charts

Slide-Ready Evaluation Visualizations

🛡️ Statutory Safeguards & Verifying Tests

🔍 What is Real vs. Simulated

🔍 Pluggable OCR Architecture for Scans & Images

🚀 Quick Start Guide

Prerequisites

1-Click Launch (Recommended)

Manual Step-by-Step Launch

👥 Demo Accounts (Synthetic Testing Credentials)

💻 Tech Stack & Repository Structure

Core Technologies

Repository Directory Tree

Running the Test Suite

🗺️ Phased Implementation Roadmap

🏆 Team AnomalyX & Statutory Disclaimer

About

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages