Feb 2025 – Present · 1 yr 4 mos · South Delhi, Delhi, India · Hybrid

• Architected a high-performance Retrieval-Augmented Generation (RAG) pipeline using Python and GCP, implementing document normalization and metadata enrichment to significantly improve retrieval accuracy.
• Designed and deployed a multi-stage GPT models(chat + embedding) , cross-encoder, re-ranker, pinecone default models pipeline for intelligent data filtering, improving analytical precision and forecasting reliability.
• Built a hybrid semantic + keyword search system in TypeScript (Bun), reducing end-to-end query latency to 0.8–1.2 seconds through optimized Promise chaining and concurrency control.
• Improved backend API performance by ~30% by optimizing AWS infrastructure (EC2, Nginx) and implementing automated CI/CD pipelines.
• Developed a scalable asynchronous data processing engine supporting 5GB parallel uploads, leveraging GCP Pub/Sub and Cloud Functions for high throughput and fault tolerance.
• Implemented CRM integrations with Zoho and HubSpot, automating lead ingestion, enrichment, bidirectional data sync, and activity tracking across platforms.
• Engineered a LinkedIn account integration and outreach automation workflow, enabling the web application to send connection requests and messages directly to prospects, with engagement status synchronized back to Zoho and HubSpot.
• Built a Unified BI Dashboard aggregating CRM, LinkedIn, and application data, featuring real-time visualizations using Recharts and automated reporting via Cron + Nodemailer.
• Eliminated LLM cold-start latency by implementing a warm-start strategy for Pinecone SDK and OpenAI embedding workflows, ensuring consistent low-latency responses.