Sitanshu Anmol

An EDA agent platform where users can upload CSVs run provided analyses tools or ask queries to dedicated chatbot for insights and answers.

Uses PostgreSQL for persistent runs and results, Redis and Celery workers for background jobs, WebSockets for real-time tool-run updates.

Python, FastAPI, Next.js, PostgreSQL, Redis, Celery, WebSockets, Cloudflare R2

A persistent memory infrastructure package for AI agents with provider-adaptive LLM and embedding integrations, and configurable vector backends.

Includes candidate-memory extraction and update/delete flows, with optional knowledge graphs for representing entity relationships.

Python, LangChain, SQLite, Neo4j

A modular multi-agent research system that converts raw topics into structured academic reports, also has retry logic, and manual task generation.

Uses Gemini and GPT-OSS models across the pipeline, with LangChain tracing and experiment monitoring.

Python, FastAPI, LangGraph, LangChain

A scaled-down Qwen3 implementation built from scratch with support for both Dense Transformer and Mixture-of-Experts language models using a shared training framework.

Implements RoPE, Grouped Query Attention, RMSNorm, SwiGLU, checkpointing, gradient clipping, and configurable GPU training.

Python, PyTorch, NumPy

A modular vector-search benchmarking library for comparing FAISS index types across recall@k, latency, and memory trade-offs on real datasets.

Supports hybrid retrieval with BM25 and dense search through reciprocal rank fusion, alongside metadata filtering and a unified retrieval interface.

Python, NumPy, FAISS, Retrieval