Hi, I'm
Anshul Pandey.
Data Scientist & AI Engineer
I architect high-performance, production-grade intelligent systems—specializing in Generative AI, low-latency RAG pipelines, and hyper-scalable recommendation engines deployed to millions of users.
Work Experience & Core Capabilities
Engineering scalable data intelligence platforms and shipping applied machine learning models into production.
GenAI & Agentic Architectures
Architecting multi-agent reasoning systems, hybrid dense/sparse vector search, context reranking, and self-hosted low-latency LLMs with 4-bit quantization.
Predictive Modeling & RecSys
Collaborative filtering, Apache Pinot integration, CTR optimization, and subscriber churn classification.
Scalable Serving & Infra
Flask/FastAPI production backends, GPU worker scaling, token streaming, and Docker orchestration.
Jr. Data Scientist
Amar Ujala Web Services Pvt. Ltd.
- Content-Based Recommendation System: Designed and deployed a platform-agnostic recommendation model in production, increasing CTR by 100%. Debugged MMR pipeline for relevance.
- Item-Based Collaborative Filtering: Built an item-item recommender extracting user events from Apache Pinot, served via Flask REST API with automated retraining.
- News Gap Detection: Built a Flask API comparing stories against an Elasticsearch news index using OpenAI embeddings, FAISS vector search, and MMR to identify missed topics.
- Subscriber Churn Prediction: Engineered engagement features and trained classification models to predict subscriber churn risk for targeted retention campaigns.
Data Science Intern
Amar Ujala Web Services Pvt. Ltd.
- NLP Chatbot Development: Developed a multi-platform NLP chatbot automating interactions, utilizing GPU-accelerated FAISS indexing for real-time semantic search.
Data Analyst (Contract)
Topper Mantra
- Product Analytics & ML Data Foundation: Designed pre-launch product analytics strategy, defined KPIs, and event-tracking taxonomy. Structured features for future recommendation and personalization systems.
Selected Production Builds
End-to-end architectures, deployed models, and open-source systems designed with performance, precision, and scale in mind.
Built a retrieval-augmented generation (RAG) chatbot chunking PDF text into Voyage embeddings stored in ChromaDB, utilizing Claude LLMs for grounded, context-aware answers.
- Anthropic Claude
- ChromaDB
- Voyage Embeddings
Deployed open-source Chatterbox voice cloning on local GPUs. Built an end-to-end Hindi speech pipeline optimized with quantization, KV-cache optimization, and token streaming for low-latency output.
- Python
- PyTorch
- Chatterbox TTS
- GPU Inference
Thoughts & Engineering Notes
In-depth articles covering Generative AI architectures, RAG optimization benchmarks, self-hosted LLM quantization, and production deployments.
Explore All Articles