Shipping

Backend ownership across APIs, performance optimization, caching/rate limiting, and security hardening.

• Reduced API latency by 25% for 10,000+ daily users by building 15+ DRF endpoints and optimizing complex PostgreSQL queries.
• Optimized Django ORM query execution to reduce unnecessary DB hits and improve response time on critical flows (query count reduction, better access patterns).
• Enabled the system to handle 5x traffic spikes during peak loads with a Redis-based sliding-window rate limiter and caching strategy, improving response times by 40%.
• Implemented strict domain whitelisting and IP-based access control middleware, significantly hardening the API security posture.
• Worked closely with product/QA/dependent teams to align functionality, confirm edge cases, and deliver features end-to-end.

Shipped production features while improving caching efficiency and database performance.

• Increased team sprint velocity by 15% by delivering 12 production-ready features within 3 months through Agile/Scrum cycles.
• Reduced direct database query load by 30% using a multi-layer Redis caching mechanism, improving cost efficiency.
• Improved maintainability by standardizing request validation, error handling, and reusable service-layer logic for APIs.
• Collaborated in reviews and debugging sessions to ship stable releases and reduce regressions.

Built end-to-end ML pipelines, improved predictive accuracy, and strengthened model training/evaluation workflows.

• Built end-to-end ML pipelines using TensorFlow and scikit-learn and achieved 85% validation accuracy on predictive models.
• Improved model performance by 20% for core business cases by fine-tuning deep learning architectures using transfer learning.
• Worked on dataset preparation and feature preprocessing to stabilize training and improve generalization across validation splits.
• Designed repeatable experimentation workflows (training + evaluation) to compare model variants and track improvements over time.
• Applied an engineering mindset to ML delivery: reproducible runs, clear metrics, and clean model interfaces for downstream usage.
• Developed evaluation framework measuring precision, recall, F1-score, and business metrics (revenue impact), enabling data-driven model selection and deployment decisions.
• Leveraged GenAI ecosystem familiarity (HuggingFace, OpenAI APIs, RAG concepts from other projects) to communicate effectively on AI-related product requirements.
• Optimized model inference using ONNX Runtime and quantization techniques, reducing model size by 60% and inference time by 40% for edge deployment scenarios.