Pushpak Rai
Software Engineer with 1+ years of experience building backend services, Generative AI applications, and Retrieval-Augmented Generation (RAG) systems using Python, FastAPI, LangChain, and Microsoft Azure. Experienced in NLP, machine learning, vector search, model evaluation, and cloud deployment.
Skilled in developing scalable APIs, document intelligence solutions, and production-ready AI workflows with a strong foundation in data preprocessing, feature engineering, and software engineering best practices.
Experience
Software Engineer
Zensar Technologies
Pune, India
Built and maintained FastAPI-based backend services for LLM orchestration and enterprise AI APIs, integrating RAG pipelines, embedding retrieval, and prompt assembly for document QA systems.
Developed data preprocessing and feature engineering pipelines using Pandas, NumPy, and scikit-learn to support NLP and AI model development.
Built document ingestion, indexing, and retrieval workflows using embeddings and vector search techniques to support AI-powered search and question-answering applications.
Used scikit-learn utilities for data preprocessing, feature engineering, experimentation, and evaluation of NLP workflows.
Implemented request orchestration flow in FastAPI with async processing and structured logging for traceability of LLM inference requests.
Deployed AI models on Microsoft Azure using Docker and CI/CD pipelines, ensuring reliable and scalable service delivery.
Implemented secure API authentication and authorization using JWT and RBAC for enterprise application security.
Implemented structured logging and request tracing to improve observability and simplify debugging of AI workflows.
Education
Godavari College of Engineering
Bachelor of Technology
CGPA: 7.1/10
Skills
Projects
RAG-Based Enterprise Document Intelligence Platform
Python, LangChain, FAISS, FastAPI, Microsoft Azure
Designed a RAG pipeline using LangChain for document retrieval, context enrichment, and multi-step question answering.
Implemented query routing and hybrid retrieval using BM25 and Vector Search to improve document relevance and response quality.
Used embedding models and FAISS vector storage for semantic retrieval, applying recursive chunking and overlap strategies to preserve context across document segments.
Applied chunk-size tuning, overlap optimization, confidence thresholding, and retrieval fallback handling to improve response grounding.
Developed FastAPI backend with async processing for concurrent inference requests, deployed on Microsoft Azure using Docker and CI/CD.
Added comprehensive logging and tracing for all LLM inference requests and tool executions to enable debugging and performance monitoring.
NLP-Powered Customer Insights Platform
Python, PyTorch, scikit-learn, BERT, PostgreSQL, FastAPI
Built supervised and unsupervised NLP workflows for sentiment analysis, intent classification, and topic clustering.
Used BERT embeddings, scikit-learn, Pandas, and NumPy for text preprocessing, feature engineering, model training, and cross-validation to build robust NLP workflows.
Exposed results through FastAPI and stored structured outputs in PostgreSQL for business intelligence.
Deployed as a containerized microservice using Docker, improving scalability and maintainability.
Certifications
Oracle Cloud Infrastructure 2025 Certified Generative AI Professional
Python for Backend Development
edX - IIT Kharagpur
AI Upskilling Certificate: Technical Foundation
Qualcomm Academy