Vetted PyTorch Professionals

“Built multiple systems from scratch at DSSD and Aglint, including an NGO sustainability reporting dashboard and a production LLM-powered phone screening agent using Twilio/Retell AI with RAG grounded in PostgreSQL candidate/job data. Strong focus on real-world reliability: guardrails, monitoring, and lightweight eval/regression loops that reduced recruiter score overrides by ~30%. Currently on OPT through May 2026 (plans STEM OPT extension) and committed to relocating to NYC for in-person work; seeking $90k–$120k base with meaningful equity for founding engineer roles.”

“LLM engineer who built and deployed an emotionally intelligent AAC communication system using an emotion-aware RAG pipeline (Empathetic Dialogues + GoEmotions) and a PEFT-adapted model. Experienced with LangChain/LangGraph and custom Python orchestration, focusing on reliability (guards, schema validation, fallbacks), latency optimization, and rigorous evaluation (automatic metrics + human-in-the-loop), with a reported 18% user satisfaction improvement.”

“Built core perception and decision layers for a 3D AI-powered interactive avatar/agent with a robotics-like perception–reasoning–action loop, combining computer vision, NLP, and real-time response. Focused on making multimodal inputs robust (normalization, intent + emotion signal fusion) and improving real-time performance via instrumentation, profiling, and parallelization; also designed distributed, loosely coupled state-based communication and deployed services with Docker.”

“Robotics software engineer working on GNSS-denied UAV localization using 5G PRS (OpenAirInterface + USRP B210) and multi-sensor fusion, with a published AIAA SciTech 2026 result achieving 1.5m RMSE on low-cost hardware. Also integrating a Vision-Language-Action model (SmolVLA) onto the Stretch 2 platform for language-assisted manipulation, leveraging ROS 2, imitation learning data collection with RGB-D, and simulation in Gazebo/MuJoCo for sim-to-real deployment.”

“Backend/ML engineer focused on clinical AI transparency who built ShifaMind, an explainability-enforced clinical ML system using UMLS/MIMIC-IV/PubMed data with RAG, GraphSAGE, and cross-attention. Demonstrated strong production engineering via FastAPI API design and safe migrations (feature flags/shadow inference), plus HIPAA-aligned auth/RLS patterns; also delivered a real-time comet detection system reaching 97.7% accuracy.”

“Robotics/AV-focused candidate who contributed to an F1TENTH autonomous vehicle college project, building key autonomy components from raw sensor data to driving commands. Strong in perception and state estimation (visual odometry, particle-filter localization), plus mapping (occupancy grids) and planning/control (RRT, Gap Follow, PID), with hands-on ROS tooling and simulation validation in Gazebo/RViz and ROS environment containerization using Docker.”

“Backend/data engineer with hands-on production experience across Azure and AWS: built FastAPI + PostgreSQL services with Azure AD OAuth2/JWT auth and strong reliability patterns (timeouts, retries, correlation IDs). Delivered AWS Lambda/ECS solutions with Terraform/CI-CD and cost controls (SQS buffering, reserved concurrency), and built/operated AWS Glue ETL pipelines into Redshift while modernizing legacy SAS reporting into Python microservices with parity testing.”

“Backend engineer with legal-tech and AI workflow experience: built JurisAI, an end-to-end legal research system using OCR + embeddings + Pinecone vector search to deliver citation-grounded LLM answers with safe failure modes (~90% recall@K). Also led a GW Law metadata migration into Caspio with batch validation and parallel rollout, and has strong FastAPI/GCP production reliability and observability practices.”

“At Trellis, built and shipped a production multi-agent, authenticated GenAI chatbot for sensitive financial account inquiries (loan/payment lookups), using dynamic model routing to control latency and cost while improving accuracy. Implemented prompt-injection defenses (Meta Prompt Guard), RAG with LangChain, and LLM-as-a-judge evaluation; the system cut manual support call volume by 40%+ and was refined through close collaboration with QA-driven user testing.”

“Built and deployed a transformer-based clinical document classification system that processes unstructured clinical notes in a HIPAA-compliant healthcare setting, served via FastAPI on AWS and integrated into an Airflow/S3 pipeline. Demonstrates strong end-to-end MLOps skills (data quality remediation, low-latency inference optimization, monitoring with MLflow/CloudWatch) and effective collaboration with clinicians to drive adoption.”

“Early-career engineer who built a self-directed production-grade security scanning/analysis pipeline that normalizes multi-scanner results, correlates CVEs, and uses an LLM to generate exploit hypotheses—then hardened it for real-world reliability (timeouts, confidence scoring, feature flags, graceful degradation). Also integrated a real-time audio ML model into Discord/Zoom and debugged intermittent latency/dropouts across Python inference, virtual audio drivers, and network jitter; experienced with IAM integrations (Entra ID/Salesforce) and cloud tooling (AWS/Docker/Kubernetes).”

“Built and deployed an LLM-to-SQL automation system in a closed/internal environment, using a retriever–reranker–validator architecture on Kubernetes with strong security controls (semantic + rule-based validation and RBAC), achieving 99% uptime and cutting manual query time ~40%. Also worked on genomic sequence classification and semantic search workflows, orchestrating data prep with Airflow, tracking/deploying with MLflow, and optimizing distributed multi-GPU training on a university Kubernetes cluster.”

“At Provenbase, built and shipped a production LLM-powered semantic search and candidate matching platform (RAG with GPT-4/Gemini, multi-agent orchestration, Elasticsearch vector search) to scale sourcing across 10M+ candidate records and 1000+ data sources. Drove sub-second performance, cut LLM spend 30% with routing/caching, and improved recruiting outcomes (+45% sourcing accuracy; +38% visibility of underrepresented talent) through bias-aware ranking and tight collaboration with recruiting stakeholders.”

“Backend/data engineer who built a production analytics data service (Python/FastAPI on AWS/Postgres with PySpark ETL) handling millions of records per day and drove major latency improvements (10–15s to <2s) via indexing, Redis caching, and shifting aggregations into ETL. Also shipped an LLM-based natural-language-to-SQL assistant end-to-end with strong guardrails (schema restrictions, read-only validation, RBAC, masking) and designed a multi-step agent workflow with verification and fallback logic.”

“Built and productionized a domain-specific LLM-powered RAG knowledge assistant at JerseyStem for answering questions over large internal document corpora, owning the full stack from FAISS retrieval and LoRA/QLoRA fine-tuning to AWS autoscaling GPU deployment. Drove measurable gains (28% accuracy lift, 25% latency reduction) and improved reliability through hybrid retrieval, grounded decoding, preference-model reranking, and Airflow-orchestrated pipelines (35% faster runtime), while partnering closely with non-technical stakeholders to define success metrics and ensure adoption.”