Systems & Architecture

Pranav Thakur

Designing high-throughput cognitive architectures, low-latency RAG pipelines, and secure agentic frameworks for real-world problem-solving.

Explore Space

About Architecture

I am a hands-on AI Systems Engineer dedicated to bridging the gap between theoretical machine learning and scalable production architectures. I specialize in designing intelligent applications that turn raw, unstructured data into high-value cognitive insights.

My development philosophy is rooted in first-principles thinking: I believe in understanding model behaviors, optimizing vector indexing paths, and crafting secure backend layers from scratch. By treating code as an engineering blueprint, I actively build systems that are modular, deterministic, and highly responsive.

Core Philosophy

First-principles engineering, deterministic output guarantees, and resource-conscious execution paradigms.

Current Research

Low-cost embeddings, localized agent routing, and hybrid semantic-lexical search indexes.

Academic Foundation

Sanjay Ghodawat University

B.Tech in Artificial Intelligence & Data Science

Focus on advanced spatial computing algorithms, mathematical logic, statistical model evaluation, and high-performance interactive media architectures.

NxtWave of Innovation in Advanced Technologies

Industry Training & Academic Collaboration

Hands-on industry-oriented training through NxtWave, focused on AI, Data Science, full-stack development, problem-solving, and modern software engineering practices through project-based learning and real-world applications.

Cognitive Tech Stack

Intelligence & Models

PyTorch

LangChain

RAG Pipelines

Vector Embeddings

LLM Fine-Tuning

Pipelines & Data

Python

Flask

SQL / SQLite

API Integration

Google Colab

Interfaces & Tools

React.js

HTML5 & CSS3

Streamlit

Git / Version Control

Render / Vercel

Featured Systems

RAG Pipeline Architecture

PDFast Engine

An ultra-fast, zero-overhead document ingestion system that optimizes context window mapping for dense vector searches.

Implemented miniLM embeddings for local indexing
Configured Groq's hardware LPU API for ultra-low latency response
Designed custom chunk partitioning algorithms

Performance: Zero-cost localized vector indexing at scale

Streamlit LangChain Groq API All-MiniLM

Generative Orchestration

Prompt2Course

A structural syllabus orchestration engine that recursively maps generalized intent prompts into logical multi-module nodes.

Constructed hierarchical routing rules for content generation
Integrated dynamic regional language translation agents
Designed asynchronous rendering interfaces to prevent request timeouts

Outcome: Generated 100+ fully-structured courses instantly

React.js LLM APIs Asynchronous Workers Vercel

Heuristic Classification

Smart Scam Detector

High-precision pattern classifier that analyzes incoming message bodies to isolate fraudulent activity and malicious formats.

Engineered heuristic parsing logic for capitalization & exclamation checks
Trained statistical models to classify vector outputs into SPAM/HAM
Achieved robust sub-second inference speeds in production

Accuracy: High-performance low-resource inference

Python Streamlit Heuristic ML Scikit-learn

State & Relational Systems

Task Blueprint Manager

Multi-user scheduling state machine featuring robust session validations and strict relational schema indexing.

Built state transition patterns for secure CRUD execution
Designed normalized SQLite database mapping for multi-tenant users
Developed category routing endpoints and dynamic prioritization filters

Security: End-to-end encrypted session authentications

Python Flask SQLite Authentication

Engineering History

Co-Head, National Service Scheme (NSS)

Sept 2025 - June 2026

Leading department-wide operational logistical workflows and coordinating large scale community welfare architectures. Directed regional team efforts to drive community engagement, managing structural communications, and organizing volunteer activities involving 100+ active participants.

Backend & AI Orchestrator @ Vibe Hack 2.0

Dec 2025

Directed a engineering team of 4 to design, build, and deploy an automated, cognitive learning assistant using Gemini 2.5 Flash. Oversaw database model schemes, vector routing parameters, and API integration paths, leading to a highly responsive and custom learning assistant within a high-pressure 48-hour timeline.

Secure Channel