Available for new opportunities

Hi, I'm Anshul Pandey.

Data Scientist & AI Engineer

I build intelligent systems, specializing in applied Generative AI, RAG pipelines, and hyper-scalable LLM architectures for production.

ResumeMailWhatsApp
Anshul Pandey
@anshul-20
Inference
10x Speed
RAG Pipeline
Synced

02.My Experience

What I Do: GenAI & LLMs

I architect intelligent agents and RAG pipelines using LangChain, LangGraph, and optimized local models via quantization.

Hugging FaceLangChain & LangGraphFAISS & ChromaDB

Machine Learning

Scikit-learn, TensorFlow, predictive analytics, and semantic clustering algorithms.

Vector DB & Infra

Linux deployment, REST APIs, GPU scaling, FAISS, and ChromaDB integration.

August 2024 — Present

Jr. Data Scientist

Amar Ujala Web Services

  • Designed LLM-powered meta-content pipelines automating news summaries and structured fact extraction.
  • Built an embedding-based feed gap analysis system clustering competitor RSS feeds via FAISS over massive datasets.
  • Slashed inference costs utilizing quantized models (4-bit/8-bit), KTransformers, and rigorous GPU batching setups.
  • Deployed 'Chatterbox' Hindi voice cloning pipeline entirely on local GPU infra, securing total data privacy and bypassing commercial APIs.

03.Featured Builds

Voice AI / Synthesis

Local Hindi Voice Cloning

An end-to-end pipeline integrating text preprocessing, phoneme alignment, and inference for high-quality Hindi speech. Handled KV caching and token streaming to optimize latency on long audio.

  • Chatterbox
  • Open-source TTS
  • GPU Quantization
  • Python

Data Pipeline / Clustering

Feed Gap Analysis Engine

Automated RSS ingestion analyzing multi-source news streams. Generates dense embeddings matched via FAISS to power a coverage scoring framework that identifies under-reported topics structurally.

  • FAISS
  • Embeddings
  • Clustering
  • Automated Batching

Thoughts & Findings

A collection of deep dives into Generative AI implementations, optimizations, and the nuances of building intelligent systems in production.

Read the Blog