Vetted FAISS Professionals

“ML/NLP practitioner with a master’s thesis focused on domain-adaptive knowledge distillation for LLMs (LLaMA2/sheared LLaMA), showing improved perplexity and ROUGE-L on biomedical data. Also built real-world data linking and search systems: integrated ClinicalTrials.gov with FAERS using fuzzy matching + embeddings, and delivered an LLM-powered FAQ recommender at Hyperledger using sentence-transformers, FAISS, and fine-tuning to mitigate embedding drift.”

“Built and shipped a production LLM pipeline for nursing home Medicare reimbursement (PDF OCR + fact extraction + keyword RAG + QA) that reportedly increased payouts by ~$1K/month per patient. Strong in LLM ops/benchmarking (ground truth, LLM-as-judge, cost/I-O tracking) and pragmatic optimization—swapped retrieval approaches, fine-tuned a small model to cut OCR cost 90%, and migrated workloads to Azure/Temporal to scale nightly processing 10x.”

“LLM/RAG engineer who built a production-style multi-agent orchestrator for resume-to-recommendation workflows (PDF ingestion through screening and recommendations), emphasizing prompt tuning and strict JSON output contracts. Currently building a RAG application for an NGO using Airflow (DAGs + embeddings) and tackling messy, missing/imbalanced data; has hands-on retrieval stack experience (FAISS/HNSW, bge embeddings) and uses rigorous evaluation metrics for groundedness and hallucination control.”

“Built and deployed a production LLM-powered risk assistant for KPMG and Freddie Mac that lets analysts query a confidential Neo4j risk graph in natural language (no Cypher), turning multi-day analysis into minutes with traceable, cited answers. Implemented rigorous guardrails, deterministic verification, RBAC/security controls, and a full eval/observability stack, cutting query error rate by ~50% and iterating through weekly UAT with non-technical risk analysts.”

“ML/LLM engineer at Adobe who deployed a transformer-based personalization and campaign-targeting recommender system end-to-end, including PySpark/Airflow pipelines processing 12M+ events/day and containerized inference on AWS SageMaker (Docker/Kubernetes). Also has hands-on LLM workflow experience (RAG, semantic search, prompt optimization, hallucination mitigation) with a metrics-driven approach to reliability, drift monitoring, and reproducible retraining via MLflow.”

“Built a production internal knowledge assistant using a RAG pipeline over large spreadsheets, PDFs, and support documents, using transformer embeddings stored in FAISS. Focused on real-world production challenges—format normalization, retrieval quality, hallucination reduction (context-only + citations), and latency—using hybrid retrieval, quantization, and containerized deployment, and communicated the workflow to non-technical stakeholders using simple analogies.”

“Built a production RAG-based natural-language-to-SQL system at Global Atlantic to replace slow, expensive manual analytics ticket workflows, focusing heavily on retrieval quality and measurable evaluation (200-question ground-truth set; recall@5 improved 0.65→0.78 via semantic chunking). Also built a custom MCP-style agent orchestrator for a personal project (arxiv-ai) to improve flexibility and Langfuse-aligned observability, and has hands-on experience with LangGraph, CrewAI, and n8n.”

“Built a production LLM-driven archival assistant that turns large, low-quality scanned handwritten files (120+ pages) into structured datasets, overcoming context-window and hierarchy challenges with a two-phase LLM + rules pipeline and reaching 98.1% accuracy (Gemini-2.5 Flash). Also orchestrated a large human-in-the-loop effort with 78 archivists, producing 2,400 high-quality annotations in 4 days via detailed rubrics and support.”

“ML/LLM engineer with production experience building a multi-GPU LLM inference platform using TensorRT and vLLM, achieving ~40% p95 latency reduction through batching/KV caching, quantization, and CUDA/runtime tuning. Also has end-to-end orchestration experience (Kubernetes, Airflow) and has delivered real-time fraud detection systems at Accenture in close collaboration with non-technical risk and product stakeholders.”

“Built a production RAG-based "knowledge copilot" for support/ops using LangChain/LangGraph, implementing the full pipeline (ingestion, chunking, embeddings, vector DB retrieval/rerank, guarded generation with citations) and operating it as monitored microservices with CI/CD. Also designed an event-driven, streaming backend for real-time inventory ordering predictions that reduced stockouts by 25%, and has hands-on incident response experience stabilizing LLM API latency/5xx spikes using Datadog/APM and resilience patterns.”