PRIYANSHU

KUMAR

Priyanshu Kumar

Designing autonomous multi-agent platforms, memory-driven RAG systems, and production AI infrastructure.

About

I design and deploy production-grade AI systems, including a live multi-agent hiring platform with autonomous agent creation, vector memory (Pinecone, ChromaDB), and real-time agent communication.

Built and shipped across AWS, Vercel, and Netlify, my systems go beyond prototypes. They are architected for scale, memory persistence, and multi-tenant interaction. My core project, Project Nexus, unifies AI agents, job automation pipelines, and an AI-powered coding evaluation environment in one architecture.

I operate across backend, memory systems, and deployment layers.

Focus Areas: Machine Learning, AI Agents, RAG Architectures, System Design.

Seeking high-impact AI/ML and System Architecture roles where technical depth matters.

Projects

Project Nexus - Multi-Agent Intelligence Architecture

Python, LangChain, FastAPI, Vector DB, RAG

  • - Built a modular multi-agent architecture with planner, research, execution, and critic agents.
  • - Implemented structured orchestration, task decomposition, and inter-agent communication.
  • - Integrated retrieval-augmented memory to enable long-context reasoning and self-correcting outputs.

Building an LLM from Scratch - GPT Style Decoder Progression

PyTorch, Transformers, Attention, Tokenization, Jupyter

  • - Built five progressive language-model notebooks, starting from token embeddings plus GELU and ending with a full decoder-only transformer stack.
  • - Implemented positional embeddings, layer normalization, tied input-output weights, causal attention, residual connections, and transformer blocks from first principles.
  • - Structured the final model as a GPT-style 12-block decoder with multi-head self-attention and autoregressive next-token generation.

Job Share Hub - Automated Student Job Platform

Next.js, Python, Selenium, BeautifulSoup, Clerk, MCP

  • - Built a centralized platform that automates LinkedIn job discovery for students.
  • - Engineered an MCP-driven scraping pipeline with role enrichment and 24h cleanup jobs.
  • - Reduced manual job-hunt effort by 10+ hours/week for active users.

Authentify (Fakefy) - Multi-Modal Fake Detection

Python, Flask, NLP, Computer Vision

  • - Built full-stack system to detect fake news, synthetic text, and deepfakes.
  • - Integrated backend APIs to connect ML models with frontend workflows.
  • - Enabled real-time inference with deployed model pipelines.

TruthScan - Deepfake Detection System

Python, TensorFlow, OpenCV

  • - Built CNN-based deepfake detector with 88% validation accuracy.
  • - Designed video frame extraction and noise reduction pipeline.
  • - Implemented end-to-end training, evaluation, and inference flow.

URL Shortener

Node.js, Express.js, MongoDB

  • - Designed scalable URL shortening and redirection APIs.
  • - Added analytics for clicks, geolocation, and device metadata.
  • - Structured modular backend for performance and maintainability.

KNOWLEDGE SPACE

Blog Articles

Building an LLM from Scratch: From Embeddings to a GPT-Style Decoder

#LLM ARCHITECTURE

Building an LLM from Scratch: From Embeddings to a GPT-Style Decoder

How I built five progressive PyTorch notebooks that evolve from a tiny embedding-plus-linear language model into a full decoder-only transformer with causal attention.

Read articleMar 12, 20269 min read
Project Nexus: Building a Multi-Agent Intelligence Architecture

#AGENTIC SYSTEMS

Project Nexus: Building a Multi-Agent Intelligence Architecture

How I engineered a modular multi-agent architecture where specialized agents collaborate, reason, retrieve memory, and self-correct.

Read articleMar 11, 202612 min read
Reclaiming Time: Why (and How) I Built Job Share Hub for Students

#PLATFORM ENGINEERING

Reclaiming Time: Why (and How) I Built Job Share Hub for Students

How I built an automated LinkedIn job intelligence platform that saves students 10+ hours weekly through scraping, enrichment, and static data delivery.

Read articleMar 10, 202610 min read
View All Blogs

Experience

Experience section will be added soon.

Education

B.Tech in Computer Science & Engineering

Pranveer Singh Institute of Technology, Kanpur (2023-2027)

CGPA: 8.5+

Senior Secondary (CBSE)

Kendriya Vidyalaya IIT Kanpur

Percentage: 95%

Contact

priyanshu.altruist@gmail.com +91 91297 46736 github.com/Priya123nshu linkedin.com/in/priyanshu-kumar-980b50179 portfolio-vert-two-73.vercel.app