SDE - 1 at The Digital Loom (2024-06 – Present)
Lead backend development for two production platforms — an AI-powered multi-tenant survey intelligence system and an enterprise ERP automation platform.
- Architected and launched Leap Survey Platform end-to-end from zero to production, owning system design, backend engineering, AI integration, and deployment
- Engineered a scalable backend using FastAPI, MongoDB, and OpenAI GPT-4o-mini to automate survey creation and response analysis
- Established a structured LLM survey generation pipeline with JSON schema validation to ensure reliable parsing of model outputs
- Reduced survey generation latency by 40–60% through parallel asynchronous execution using asyncio.gather
- Crafted an AI analytics pipeline converting raw survey responses into statistical summaries and GPT-generated narrative reports
- Introduced secure enterprise authentication using JWT, bcrypt, and hierarchical RBAC enabling role-based access across teams
- Containerized backend services with Docker and Docker Compose, enabling streamlined deployments
- Enforced strict multi-tenant data isolation across enterprises, teams, surveys, and responses
- Devised and deployed an ERP automation solution using Microsoft Power Platform to digitize operational workflows
- Produced 5 model-driven applications and 20+ automation flows replacing manual processes
- Reduced manual workload by 12+ hours per month and decreased operational errors by 90%
- Generated operational alerts producing 500+ notifications monthly
- Led multi-environment deployments ensuring reliable adoption across teams
AI & ML Intern at Khanuja Group (2023-12 – 2024-06)
Developed AI and machine learning solutions including automated face-recognition systems.
- Created an automated face-recognition attendance system using MTCNN, FaceNet, VGG16, and ResNet
- Developed a complete computer vision pipeline integrating detection, embedding generation, and identity recognition
- Achieved 98% identification accuracy and reduced manual attendance tracking time by 80%
