Research Agents
Multi-agent scientific systems that ingest papers, extract structured knowledge, and generate hypotheses.
Ahmedabad based engineer | B.Tech IT | AI and full-stack systems
Vivek Jadav
I build AI systems that move from research notebooks to useful products: RAG APIs, scientific agents, small-transformer experiments, and full-stack platforms with real users.
Grounded RAG
Hybrid retrieval, FAISS/pgvector indexes, local LLM generation, and cited answers over domain documents.
Product Systems
Next.js, FastAPI, Prisma, PostgreSQL, MongoDB, Docker, RBAC, dashboards, and deployable user flows.
Selected work
Featured Projects
Chosen from my public GitHub and resume for depth, recency, and product value.
Research agent
2025 - Present
Scientific Copilot
A self-improving scientific research agent. It fetches and validates ArXiv papers, stores paper atoms in PostgreSQL with pgvector, and supports a multi-agent hypothesis loop.
Enterprise platform
2026
Document Management System
A secure full-stack document platform with JWT sessions, RBAC, signed uploads, version history, branch workflows, approvals, audit logs, PostgreSQL, and S3-compatible storage.
Transformer research
2024 - Present
Memorization vs. Generalization
Trained a 50M-parameter decoder-only transformer on synthetic reasoning tasks and built an evaluation suite that exposed memorization despite near-perfect in-distribution results.
RAG API
2024
RetrievalStack
A local RAG assistant for 19 reinforcement-learning papers with PDF ingestion, FAISS search, IDF-weighted reranking, TinyLlama generation, FastAPI endpoints, and Docker support.
Learning platform
2024
AI Tutor
An interactive study assistant with textbook-grounded chat, voice input, test generation, chat history, and a separate RAG engine for AI answers.
Civic product
2024
Voter's App
A secure election platform serving 200+ users with Ranked Choice and Preferential voting, real-time results, multi-organization RBAC, and GenAI-assisted content workflows.
From scratch ML
2026
NeetCode GPT
A learning-first GPT implementation with attention heads, multi-head attention, transformer blocks, tokenizers, KV cache, grouped-query attention, training, and generation.
Applied ML
2025
Fake News Detector
A Streamlit NLP app that classifies news text as real or fake using TF-IDF features, Logistic Regression, confidence scores, and a labeled Kaggle dataset.
Experience
Production work, education, and leadership
A practical engineering profile: ship the product, learn the theory, then explain it clearly.
Full-Stack Developer Intern | AUM Imagineering
Engineering a document management platform end to end: schema design, frontend flows, backend services, versioned storage, retrieval, RBAC, and deployment.
B.Tech Information Technology | GEC Bhavnagar
CGPA 8.51/10 with focus areas in DSA, DBMS, linear algebra, probability, machine learning, and deep learning systems.
President | Literary Society
Rebuilt a dormant chapter to 15 active members and organized 4 events for 200+ participants.
Research voice
What I'm exploring
Public LinkedIn snippets and GitHub work show a clear theme: AI systems are useful only when they retrieve well, reason honestly, and survive real evaluation.
High training accuracy is not the same as reasoning.
The small-transformer study reports near-perfect in-distribution accuracy but complete exact-match failure on 7,137 out-of-distribution examples.
Research tooling needs memory, structure, and feedback.
The Scientific Copilot direction turns paper overload into a structured system for ingestion, clustering, hypothesis generation, critique, and iteration.
RAG gets practical when it has a real user workflow.
AI Tutor and RetrievalStack connect retrieval pipelines to student questions, RL papers, voice interaction, test generation, and cited answers.
Stack
Tools I use to ship
AI and ML
PyTorch, HuggingFace Transformers, LangChain, LangGraph, RAG, LoRA, Whisper, FAISS, Ollama
Backend and Web
Python, FastAPI, Flask, Node.js, Next.js 14/16, TypeScript, React, Remix, Tailwind CSS
Data and DevOps
PostgreSQL, pgvector, MongoDB, Prisma ORM, MinIO/S3 storage, Docker, Git, JWT, Zod
Available for AI, backend, and full-stack roles