Vetted Retrieval-Augmented Generation Professionals

“Built and deployed a scalable, production-ready LLM knowledge assistant using a RAG architecture (LangChain + vector store/FAISS) to replace keyword search for internal documents. Demonstrates hands-on expertise in hallucination reduction and retrieval quality improvements through semantic chunking, similarity tuning, prompt design, and human-in-the-loop validation, plus strong stakeholder communication via demos and visual explanations.”

“Built a production multi-agent cybersecurity defense simulator orchestrated with CrewAI, combining Red/Blue team LLM agents, a RAG runbook retriever, and an RL remediation agent trained via state-space simplification and reward shaping for rapid incident response. Also partnered with quant analysts and fund managers to deliver an automated trading and portfolio management system using statistical methods plus CNN/LSTM models, reporting up to 15% weekly ROI.”

“Built a production "Mini RAG Assistant" for internal document Q&A, focusing on grounded answers (anti-hallucination), retrieval quality, and latency/cost optimization. Uses LangChain/LangGraph for orchestration and applies a metrics-driven evaluation loop (including reranking and semantic chunking improvements) while collaborating closely with product stakeholders.”

“Master’s program project lead who built and deployed a real-time sound recognition system (Flask + React Native + ML) that was adopted by 200+ university students. Demonstrates strong production engineering and cross-layer debugging—solving latency, unreliable uploads, and observability gaps using microservice separation, chunked/idempotent transfers, and packet-capture-driven network diagnosis—plus AWS/on-prem and IoT edge-to-cloud integration experience.”

“Built and deployed a production LLM-powered RAG knowledge system to unify operational/policy information across PDFs, wikis, and databases, emphasizing auditability and low-latency/cost performance. Improved answer relevance at scale by moving from pure vector search to hybrid retrieval with metadata filtering and reranking, and partnered closely with healthcare operations/compliance to define acceptance criteria and human-in-the-loop guardrails.”

“Accenture engineer who built and shipped a production RAG-based automation/chatbot for SAP incident triage and troubleshooting, embedding thousands of runbooks/logs/tickets into a semantic search pipeline and integrating it into Teams/Slack. Reported major productivity gains (30–60% time reduction), >90% validated answer accuracy, and sub-2-second responses, with strong orchestration (Airflow/Prefect/LangGraph) and reliability practices (guardrails, testing, monitoring).”

“At Liberty Mutual, built a production underwriting decision assistant combining LLM reasoning with quantitative models and strong auditability. Implemented a claims-based response verification pipeline that cut hallucinations from 18% to 3% and materially improved user trust/validation scores. Experienced orchestrating ML/LLM workflows end-to-end with Airflow, Kubeflow Pipelines, and Jenkins, including SLA-focused pipeline hardening.”

“Backend Python engineer who architected an event-driven order integration engine connecting EDI vendors to ERP/WMS/3PL systems, including a canonical order model and adapter framework to eliminate per-customer hardcoding. Has hands-on Kubernetes production experience (microservices, Celery workers, CronJobs, HPAs) and implemented GitOps/CI-CD using GitHub Actions, Docker, and ArgoCD, including moving deployments from on-prem to Azure.”

“Senior Data Scientist in consulting who has built production RAG systems for insurance/annuity document search at large scale (100K+ PDF pages), emphasizing grounded answers, guardrails, and low-latency retrieval. Experienced in end-to-end MLOps for LLM apps—monitoring, evaluation sets, drift handling, and safe rollouts—and in orchestrating complex pipelines with Prefect/Airflow and deploying services on Kubernetes.”

“Built a production AI pipeline at EyCrowd to automatically grade shaky outdoor user-submitted brand videos using CV + CLIP/BLIP and a LangChain RAG layer per brand, with GPT-4 generating structured JSON explanations and grades. Optimized for latency and cost (batch PyTorch inference, caching), cutting review time from ~8 minutes to <2 minutes while reaching ~90% alignment with human graders and supporting thousands of videos/day.”

“LLM/RAG-focused engineer who repeatedly takes agentic workflows from impressive demos to dependable production using rigorous evals, SLOs, and deep observability. Has led high-impact incident mitigation (22-minute MTTR during a major sale) and developer enablement workshops, and partnered with sales to close a $410k ARR enterprise deal with a tailored RAG pilot (FastAPI/pgvector/Okta/InfoSec-ready).”

“ML/NLP engineer with banking domain experience who built a GenAI-powered fraud detection and risk intelligence system at Origin Bank, combining RAG (LangChain + FAISS), fine-tuned BERT NER, and GPT-4/Sentence-BERT embeddings. Delivered measurable impact (25% higher fraud detection accuracy, 40% less manual review) and emphasizes production-grade pipelines on AWS SageMaker/Airflow with strong data validation and scalable PySpark processing.”

“Built an "Offline Study Assistant" that runs LLM inference locally on a 5-year-old Android device using Llama.cpp and the Android NDK, achieving a 27x speedup and cutting time-to-first-token from 11 minutes to 30 seconds. Also has applied backend/API experience with FastAPI, Supabase (Auth + RLS), and production hardening of a RAG system at Hashmint using Celery and Redis to eliminate PDF-processing-related query failures.”

“Backend engineer who built an Emergency Alert System with Virginia Tech for the City of Alexandria, focusing on real-time ingestion, secure dashboards, and AI-assisted prioritization. Emphasizes high-stakes reliability with guardrails (hybrid rules+LLM, confidence-based fallbacks), scalable async processing, and defense-in-depth security (JWT/RBAC plus database row-level security).”

“Backend/AI engineer who has shipped production LLM systems end-to-end, including an AI request-routing service (FastAPI + BART MNLI + OpenAI/Gemini) that improved accuracy ~25% after launch via eval-driven prompt/category iteration. Also built an enterprise document intelligence/RAG platform on Azure (Blob/SharePoint/Teams ingestion, OCR/NLP chunking, embeddings in Azure Cognitive Search) with PII guardrails (Presidio), confidence gating, and scalable event-driven pipelines handling millions of documents.”

“Research assistant at St. Louis University who built and deployed a production document-intelligence RAG system (Python/TensorFlow, vector DB, FastAPI) on AWS, focusing on grounding to reduce hallucinations and latency optimization via caching/async/batching. Also developed a personalized recommendation system for the Frenzy social platform and partnered closely with product/UX to define metrics and iterate on hybrid recommenders and cold-start handling.”

“Built a production LLM/RAG document extraction and game/quiz content workflow using LLaMA 2, LangChain/LangGraph, and FAISS, achieving ~94% accuracy and reducing turnaround from hours to minutes. Demonstrates strong applied MLOps/orchestration (CI/CD, MLflow, Databricks/PySpark), robust handling of noisy/variable document layouts (layout chunking + OCR fallbacks), and practical reliability practices (human-in-the-loop routing, drift monitoring, A/B testing).”

“Early-career backend engineer and solo builder of FrameFindr, an AI/OCR-based marathon photo tagging product used at live events. Demonstrated pragmatic scaling under tight infrastructure constraints (2GB VPS) and hands-on ownership of architecture, API design, auth (Google OAuth/JWT), and a MongoDB-to-MySQL migration with data-integrity safeguards.”

“Stony Brook Fall 2024 capstone contributor who built a ROS2-based warehouse mobile robot prototype, owning perception and SLAM integration end-to-end. Strong in real-time robotics optimization on Jetson Orin (TensorRT/CUDA, ROS2 tracing/Nsight) and in distributed ROS2 communications (DDS discovery/QoS, MAVLink-to-ROS2 bridging), with a full simulation/testing/deployment toolchain (Gazebo, CI tests, Docker/K3s).”

“Backend/AI engineer with internship experience building an AI-powered financial insights platform (FastAPI, Redis, BigQuery) and prior HCL experience leading a monolith-to-microservices refactor (Flask, Kafka) using blue-green deployments. Demonstrates strong performance/security focus (OAuth/JWT/RBAC, encryption) and measurable impact on latency, downtime, and ML model reliability; MVP was submitted to Google’s accelerator program.”

“AI/LLM engineer who has shipped production RAG chatbots using LangChain/OpenAI with FAISS and FastAPI, focusing on real-world constraints like context windows, concurrency, and latency (reported ~40% latency reduction and <2s average response). Experienced orchestrating AI pipelines with Celery and fault-tolerant long-running workflows with Temporal, and has applied NLP model tradeoff testing (Word2Vec vs BERT) to drive measurable accuracy gains.”

“Built and shipped an agentic RAG chatbot module for NexaCLM to answer questions across large volumes of contracts while minimizing hallucinations and incorrect legal interpretations. Implemented routing between vector retrieval and ReAct-style agent retrieval plus an automated grading/validation layer (cosine-similarity thresholds, retries) and deployed via GitHub Actions to Azure Container Apps, partnering closely with legal stakeholders to define risk/clause-focused objectives.”

“Built and productionized an agentic RAG assistant using Ollama + LangChain + MCP + ChromaDB to speed up and standardize access to operational knowledge from tickets and runbooks. Focused on real-world reliability: mitigated timeouts/latency with retries and concurrency limits, improved retrieval via chunking/embedding iteration, and reduced hallucinations through citation-grounding and confidence-based abstention. Also partnered with non-technical ops staff to deliver anomaly detection/monitoring by translating operational needs into model signals, thresholds, and alerting logic.”