█████╗ ██╗   ██╗██╗██╗  ██╗██╗   ██╗███╗   ███╗ █████╗ ██████╗ 
██╔══██╗██║   ██║██║██║ ██╔╝██║   ██║████╗ ████║██╔══██╗██╔══██╗
███████║██║   ██║██║█████╔╝ ██║   ██║██╔████╔██║███████║██████╔╝
██╔══██║╚██╗ ██╔╝██║██╔═██╗ ██║   ██║██║╚██╔╝██║██╔══██║██╔══██╗
██║  ██║ ╚████╔╝ ██║██║  ██╗╚██████╔╝██║ ╚═╝ ██║██║  ██║██║  ██║
╚═╝  ╚═╝  ╚═══╝  ╚═╝╚═╝  ╚═╝ ╚═════╝ ╚═╝     ╚═╝╚═╝  ╚═╝╚═╝  ╚═╝

class Avikumar:
    name       = "Avikumar Talaviya"
    role       = "Data & AI/ML Engineer"
    focus      = ["Data Engineering", "MLOps", "RAG", "LLM Applications", "Analytics"]
    building   = "Local-first document intelligence with hybrid retrieval and cited answers"
    open_for   = ["Full-time Roles", "Collaborations", "Open Source", "Freelance Projects"]

I'm a hands-on data and AI/ML engineer who turns raw data and applied research into dependable products. I work across the lifecycle: data ingestion and modeling, experimentation, retrieval and inference, API development, deployment, and monitoring. My recent work spans local-first RAG, LLM-powered applications, recommender systems, healthcare analytics, and reproducible MLOps pipelines.

AI / ML

Data Engineering & Analytics

Backend, MLOps & Cloud

LLMs & Inference

Local-first personal document intelligence with grounded, cited answers

FastAPI · React · PostgreSQL · Qdrant · Hybrid Search · Cerebras · Docker

A privacy-minded RAG application that ingests PDF, DOCX, TXT, and Markdown files, identifies people, and answers questions with expandable citations. It combines BM25-style lexical retrieval with local embeddings, reciprocal rank fusion, durable PostgreSQL storage, and recoverable vector indexing.

AI-powered career intelligence platform

Next.js · TypeScript · FastAPI · Cerebras · Supabase · Vercel

Extracts PDF, DOCX, and TXT resumes to generate job-match scores, skill-gap analysis, and tailored improvements. The production deployment separates the frontend and API, uses server-side database access, and includes CI/CD workflows.

A working GPT assembled from neural-network fundamentals

Python · PyTorch · BPE · Self-Attention · Transformers · KV Cache

Implements the path from gradient descent, backpropagation, and multilayer perceptrons to tokenization, multi-head attention, transformer blocks, training, and text generation—including grouped-query attention and KV caching.

Collaborative filtering enhanced with generative AI

FastAPI · Streamlit · SVD · K-Means · LLMs · Docker

Combines item-based collaborative filtering, query expansion, clustering, and LLM-generated descriptions to retrieve and rerank relevant books. Includes an API, interactive UI, feedback flow, and evaluation plan.

Disease-risk modeling across 100,000 health and lifestyle records

Python · SQL · Pandas · Scikit-learn · XGBoost · Statistical Analysis

An end-to-end analytics project covering data preparation, class-imbalance handling, cross-validation, model comparison, and clinically relevant evaluation using recall, F1, and ROC-AUC.

Reproducible training and serving for NYC taxi-duration prediction

Python · Prefect · MLflow · FastAPI · Scikit-learn · Docker

Builds a repeatable workflow for data loading, feature engineering, model training, experiment tracking, artifact management, orchestration, and REST-based inference.

• 🔍 Local-first RAG — hybrid retrieval, rank fusion, citations, and privacy-aware inference
• 🧱 Reliable data platforms — durable storage, reproducible pipelines, and data quality
• ⚙️ Production MLOps — experiment tracking, orchestration, CI/CD, and model serving
• 🧠 LLM systems — transformer internals, efficient inference, evaluation, and observability