Sarthak Khandelwal

Product Embeddings Platform: Trend Management Console (Python, FastAPI, Qdrant, LangGraph, FastMCP) Redesigned data schema and migrated from PostgreSQL to Qdrant, reducing trend retrieval latency from 5s to 200ms. Built a Chat Agent using LangGraph to manage and visualize product trends through natural language commands. Implemented a Model Context Protocol (MCP) server using FastMCP to standardize agent tool usage and context retrieval. Retail Planning Platform: Analytics & Forecasting (Python, FastAPI, ClickHouse, Authentik) Migrated analytics database from PostgreSQL to ClickHouse, reducing dashboard query latency from 15s to sub-500ms. Implemented RBAC and ABAC authorization flows using Authentik to secure access to sensitive sales forecast data. Developed high-performance FastAPI endpoints to serve real-time visualization data for the retail planning dashboard. Search Backend: High-throughput eCommerce Search (Python, FastAPI, Qdrant, Redis, Kubernetes, Docker) Designed and implemented a search backend microservice for eCommerce using FastAPI and Python to serve relevance-ranked results. Integrated Qdrant for vector search and Redis for caching to achieve low-latency retrieval. Containerized with Docker and deployed on Kubernetes; sustained 1000 QPS with 200ms average response time.

Smart Assist: AI-Powered Call Support (Python, FastAPI, ECS, ALB, Docker, GitLab, OpenAI, Twilio, WebSockets, PostgreSQL) Built an AI-Powered call support system with OTP verification and human fallback, handling 100+ concurrent sessions with sub-500ms latency. Integrated FastAPI, Twilio Voice, WebSockets, and OpenAI streaming for real-time transcription and intent recognition, backed by PostgreSQL. Deployed a scalable WebSocket service on AWS ECS behind ALB, containerized with Docker and automated via GitLab CI/CD.

3D Reconstruction Pipeline for Heritage Site Modeling (Python, PyTorch, AWS EC2, Docker, GitLab) Developed a scalable 3D reconstruction pipeline using Gaussian Splatting and PyTorch on AWS EC2, automating workflows via GitLab/Docker to process 1000+ images and reduce costs by 50%.

LOKAL (Location Based Dealer Finder) (Python, SQL, MongoDB, Flask, Docker, Kubernetes, Jenkins) Migrated spatial querying from in-memory KDTree to MongoDB/SQL geospatial indexing, increasing service responsiveness by 20%. Developed a geolocation-based Flask microservice to optimize dealer availability searches within specified distances. Streamlined CI/CD with Jenkins and deployed containerized services using Docker and Kubernetes for scalable orchestration. Retrieval-Augmented Generation (RAG) for Design Documents (Python, Docker, VectorDB, AWS S3, AWS Lambda) Implemented a LangChain-based RAG tool with LLMs for role-based access control and retrieval of 1000+ documents. Reduced employee document search time by up to 50% with an LLM-powered chatbot for fast, accurate retrieval. Built an ETL pipeline microservice to ingest, process, store, and manage 1000+ documents using Python, S3, Lambda, and Pinecone. Deployed the system using Terraform, Docker, Kubernetes, and Jenkins with CI/CD pipelines for scalable and reliable operations.

Migrated data and transformed queries from SQL to Snowflake.

Built responsive Art Gallery website with React/JavaScript, integrated Razorpay payment gateway and ShipRocket shipping service.