Vetted Large Language Models Professionals

“Built and productionized a MediCloud/Medicoud LLM microservice platform that lets clinicians query medical data in natural language, orchestrating multi-step RAG-style workflows with LangChain and evaluating/debugging with LangSmith. Delivered measurable gains (consistency ~70%→90% / +20%; latency ~2.0s→1.1s / -40%) by implementing structured prompts, fallback logic across multiple LLMs, hybrid retrieval tuning, and AWS Lambda performance optimizations (package size, async, caching).”

“Built and deployed a production bilingual (Bengali/English) AI virtual assistant that replaced IVR for telecom customer service at massive scale (~15M users), integrating ASR/TTS, Rasa dialogue management, and custom NLP. Overcame low-resource Bengali data and noisy call-center audio with synthetic data augmentation and transformer fine-tuning, achieving significant production gains including ~50% reduction in support calls.”

“Built and productionized an LLM-based system that summarizes large volumes of unstructured content (customer feedback/internal docs) to reduce manual analysis and surface decision-ready insights. Brings strong reliability practices—prompt/schema constraints, validation checks, orchestration with Airflow/Databricks, and rigorous component + end-to-end testing—plus experience partnering closely with business stakeholders to drive adoption.”

“Backend engineer who built and scaled AmazonProAI, a multi-tenant SaaS platform for Amazon sellers, using a modular Django/DRF monolith with strict seller-level isolation and security controls. Led a controlled SQLite-to-PostgreSQL migration and hardened bulk Excel ingestion with idempotency and data integrity constraints to prevent duplicate metrics and noisy alerts while keeping the system ready for future service extraction.”

“Early-career/learning-stage candidate focused on LLM systems; has not yet built or deployed production AI applications but is actively learning orchestration (Microsoft Semantic Kernel) and core patterns like RAG, embeddings, and model selection based on business requirements.”

“Built a production LLM-powered interview-prep app that ingests job postings and generates tailored preparation plans. Iterated from a single generalist LLM to a multi-LLM pipeline and used RAG to ground the final chat assistant on locally stored intermediate outputs; has also experimented with n8n vs Python-coded pipelines for orchestration.”

“Currently working at CBS Lab in Austria, where they implemented/replicated the "Open World Grasping" research pipeline end-to-end. Built a ROS-based RGB-D perception-to-action system using SAM 2.1 segmentation and MoveIt motion planning to generate grasp poses and execute pick-and-place/sorting with a robotic arm.”

“Built and shipped a production-grade RAG-powered news summarization and Q&A product, tackling real-world issues like retrieval drift, hallucinations, latency, and autoscaling deployment (Docker + FastAPI + Streamlit Cloud). Experienced in end-to-end ML/LLM workflow automation using Airflow, Kubeflow Pipelines, and MLflow, and has demonstrated business impact (40% inference precision improvement) through close collaboration with non-technical stakeholders at Evoastra Ventures.”