Vetted LoRA Professionals

“Backend engineer who built and migrated a large-scale document intelligence platform used by legal, healthcare, and insurance clients, processing millions of pages. Experienced moving from a monolithic, LLM-heavy approach to a modular FastAPI service architecture with ML classification + RAG, strong validation/auditability, and enterprise security (JWT/OAuth, RBAC, PostgreSQL RLS) with zero-downtime incremental rollouts.”

“Backend/platform engineer who has owned a Python/FastAPI results API and deployed it on Kubernetes with Helm and GitHub Actions-driven CI/CD. Demonstrates strong production operations mindset across performance tuning, monitoring, safe rollouts/rollbacks, and phased migrations, plus hands-on Kafka streaming experience focused on ordering and idempotency.”

“ML/NLP practitioner who built a retrieval-augmented generation (RAG) system for large financial and operational document sets using Sentence-Transformers (all-mpnet-base-v2) and a vector DB (e.g., Pinecone), with a strong focus on retrieval evaluation and chunking strategy optimization. Experienced in entity resolution (rules + embedding similarity with type-specific thresholds) and in productionizing scalable Python data workflows using Airflow/Dagster and Spark.”

“Built an AI lending assistant (RAG + DeBERTa) used by credit analysts to retrieve policies and past loan decisions, tackling real production issues like hallucinations, document quality, and sub-second latency. Deployed a modular, Dockerized AWS architecture (ECS/EMR + load balancer) with load testing, caching/precomputed embeddings, and CloudWatch monitoring, and used Airflow to automate scheduled data/embedding/vector DB refresh pipelines with retries and alerts.”

“Recent J&J intern who built a conversational RAG agent and led a shift from a monolithic model to a modular RAG workflow, cutting response time from several days to under a second by tackling data fragmentation, context retention, and embedding/latency optimization. Also worked on a large (7B-parameter) multimodal VQA pipeline for healthcare research and stays current via NeurIPS/ICLR and open-source contributions.”

“ML/NLP engineer with recent Scotiabank experience building production-grade indexing automation over large-scale emails and customer databases, combining LLM fine-tuning (Mistral, XLM-R) with fuzzy matching to exceed 95% accuracy under strict banking constraints. Also built a RAG-based chat agent using Gecko embeddings, Vertex AI Search, Gemini, and cross-encoder reranking, and delivered a text-to-SQL chatbot at SOTI through iterative fine-tuning and benchmark-driven experimentation.”

“Built and deployed an end-to-end product that brings a research-paper approach into production for large-scale time-series clustering, with attention to partitioning, latency, and scalability. Also designed a Python-based backend validation service (comparing outputs to database ground truths) and handled production reliability issues by reproducing dataset-specific crashes and hardening corner-case behavior with client-friendly errors.”

“GenAI engineer who has deployed production LLM/RAG chatbots for internal document search, focusing on reliability (hallucination reduction via prompt guardrails + retrieval filtering) and performance (latency improvements via caching). Experienced with LangChain/LangGraph orchestration for multi-step agent workflows and iterates using monitoring/logs and benchmark-driven evaluation while partnering closely with product and business teams.”

“Built production LLM + hybrid RAG and multi-agent orchestration systems at Wells Fargo to automate complaint document/audio transcript understanding and categorization, addressing vocabulary drift via embedding + vector index updates instead of frequent retraining. Strong in LLM workflow reliability (testing/benchmarks/observability) and stakeholder-facing delivery with explainability (citations/SHAP-style justifications) and Tableau dashboards.”

“Built and shipped a production RAG-based assistant that lets parents ask natural-language questions about their child’s learning progress, using pgvector retrieval (child-id filtered) and Redis caching to hit ~180ms latency. Implemented real-world guardrails and compliance (Llama Guard, COPPA, retrieval thresholds, fallbacks) with 99.5% uptime, and ran human-in-the-loop eval loops that improved satisfaction from 3.8 to 4.2 while serving 60k+ monthly users and reducing costs significantly.”