Rizwana Shaik

About

Internship project building and deploying a LLaMA-based, RAG-enabled copilot inside a Professional Services Automation platform, enabling natural-language navigation, text-to-SQL reporting, and project/resource/budget insights across multiple modules. Addressed real production issues like context drift and vague queries with hybrid search, metadata enrichment, and an intent classification/rewriting layer, orchestrated via Apache Airflow—ultimately cutting PMO reporting time by 40%.

Experience

Education

Key Strengths

• Built and deployed an LLM copilot in a PSA platform across multiple modules (Projects, Clients, Vendors, Resources, Time entries)
• Designed text-to-SQL + RAG system returning grounded explanations and charts from database data
• Solved context drift at scale using metadata enrichment and hybrid search to improve relevance
• Improved handling of vague user queries via intent detection/query classification and structured task rewriting
• Production-minded agent design: success metrics, structured schemas, offline evals with real queries + human scoring, and monitoring (latency/error rates)
• Implemented safe operation patterns: grounding, RBAC, and fallbacks to reduce hallucinations
• Used Apache Airflow to orchestrate ingestion, embedding refresh, vector store maintenance, and cache warmup for real-time copilot features
• Collaborated with non-technical PMO stakeholder; reduced reporting time by 40% through iterative feedback and training
• Designed and built Flask microservices for a PSA platform (analytics, resource allocation, AI copilot)
• Performance tuning with indexing, normalization, selective loading, batching, and caching
• Reliability engineering for microservices (health checks, retries/timeouts, load balancing, horizontal scaling, centralized logging)
• Strong SQL performance workflow using PostgreSQL EXPLAIN ANALYZE and ongoing monitoring via dashboards
• Production AI integration using LLM APIs with RAG, embeddings/vector DB, and async processing to manage latency
• Multi-tenant data isolation implementation (row-level tenant ID enforcement, schema-per-tenant, automated tests to prevent cross-tenant leakage)
• High-throughput background task optimization using Celery with monitoring/alerting (Flower, structured logs, metrics)