I'm building production-grade Generative AI systems, including RAG pipelines, LangGraph orchestration, vector search, and SQL reasoning. At NETSOL Technologies Pakistan, I engineered a hybrid chatbot that routes SQL, RAG, direct LLM, and partner queries.
I built its knowledge-ingestion pipeline with web scraping, semantic chunking, header-enriched embeddings, ChromaDB retrieval, and cross-encoder reranking, reaching near 39/40 correct answers on validated queries. I also implemented authentication, RBAC, checkpointing, SQL validation, caching, rate limiting, and read-only database access.
My research evaluates adversarial robustness in small language models. Across four sub-1B models, six attacks, 170 Q&A pairs, and 9,520 inference passes, I identified tokenizer-collapse and prompt-injection vulnerabilities using EM, F1, perplexity, and attack-success-rate evaluation.
I've also fine-tuned Qwen 2.5B with LoRA for a financial AI assistant, developed a LangChain RAG pipeline, designed a PostgreSQL logistics data warehouse, and built computer-vision lane detection and tracking systems.
