M.Sc. Thesis Project • 2026

VidyaNetra AI

Stop feeding AI garbage. Our trust-aware gatekeeper dynamically audits the internet's credibility, ensuring your research is backed by undeniable truth, not SEO noise.

How it Works

The "Garbage In, Garbage Out" Paradox.

Standard Retrieval-Augmented Generation (RAG) models suffer from a fatal flaw: they fetch data based purely on semantic similarity. They blindly ingest SEO spam, unmoderated forums, and unverified blogs.

If the retrieved context is false, the AI will confidently synthesize a lie. VidyaNetra replaces semantic search with epistemic search.

Naive RAG Architecture

Faithfulness Score: 0.61

VidyaNetra Trust Architecture

Faithfulness Score: 0.94

Empirical Performance

Statistically significant results proven via the RAGAS mathematical evaluation framework.

55.9%

Improvement in RAGAS Faithfulness

3.6s

Median Time-to-First-Token (TTFT)

81%

Token Payload Reduction via Stripping

The Epistemic Pipeline

How VidyaNetra transforms chaotic web noise into verified academic truth.

1

Parallel Extraction

Deep web scraping using asynchronous Promise.allSettled() to concurrently fetch unstructured HTML data from global Serper APIs.

2

Trust Filtration

The algorithm drops untrusted forums and grades domains mathematically. Only sources passing the strict threshold survive the purge.

3

Generative Synthesis

Google Gemini 2.5 Flash ingests ONLY the verified context window to generate a flawless, hallucination-free response.

System Architecture

Domain Heuristics

Maximal weights applied to .gov, .edu, and verified institutional domains while penalizing unmoderated public forums.

Stateless Routing

Deployed on the Vercel Edge Network. Next.js API endpoints provide horizontal scalability without heavy monolith servers.

Algorithmic Transparency

Eliminates the "black box" effect. The UI directly exposes the backend $T_{final}$ mathematical score via distinct Trust Badges.

Safe Failure Execution

If the Trust Engine cannot find mathematically verified data, it executes a programmatic "Safe Failure" rather than fabricating a lie.

Cosine Similarity Vectors

Content is embedded and ranked against the user query using high-dimensional cosine similarity to ensure extreme topical relevance.

Cross-Platform Readiness

The decoupled Next.js API architecture allows seamless future integration into React Native mobile applications.

Get in Touch

Interested in discussing the VidyaNetra AI architecture, reviewing the thesis research, or exploring collaboration opportunities?