Vinay MendoleAI Engineer
SYSTEM ARCHIVE // AGENT ENDPOINT

Vinay Mendole

AI/ML Engineer & Systems Developer at National Institute of Technology Delhi. Specialized in Reinforcement Learning, NLP architectures, and full-stack systems.

// MACHINE-READABLE ENDPOINTS:
• Raw Schema: /.well-known/ai-agent.json
• AI Markdown: /?ai=true
• RSS Syndication: /rss (Pending)
• GitHub: github.com/topspeed69
FULL BIOGRAPHY →DOWNLOAD RESUME →
← SELECTED PROJECTS SPECIFICATION SHEETS & BLUEPRINTS
SPECIFICATION SHEET // REF-02
FIRST COMMIT: 2024-08-20·TAG: v1.4.0 · STABLE

Legal AI Assistant: Indian Legal Corpus Intelligence Substrate

Domain-adapted legal intelligence system with semantic statute linking and citation verification.

AUTHORVinay Mendole
LANGUAGEPython / PyTorch / ONNX Runtime
TOPOLOGYHierarchical Transformer Attention + Milvus Vector Mesh
LICENSEMIT

¶ 01Context & Design Rationale

“Legal prose is dense, hyper-referential, and adversarial. Generic language models routinely hallucinate section numbers or misunderstand jurisdictional hierarchies.”

By fine-tuning domain-specific encoders (InlegalBERT) alongside instruction-tuned decoder architectures, this engine structures Indian Supreme Court and High Court rulings into verifiable statutory citations. Every generated summary links back to exact paragraph offsets in original law reports.

¶ 02Architectural Pipeline

SCALE: LEGAL CORPUS
DOCUMENT PARSER
PDF / Text Normalization
Clean Layout Extraction
→
INLEGALBERT
Clause Segmentation
Attention Masks
→
STATUTE LINKER
IPC / CrPC Knowledge Graph
Exact Section Index
→
SYNTHESIS CORE
T5 & Gemma Fine-Tuned
Citation Verification
FIG 2.1 — LEGAL ATTENTION CLUSTER AND STATUTE DISAMBIGUATIONP95 SUMMARY LATENCY < 450ms
<> VIEW SOURCE ON GITHUB ↗