Senior Agentic AI Engineer

ANMOL TRIPATHI

AI Engineer LLM / RAG / Agentic AI (LangGraph, MCP) Machine Learning & Deep Learning Fort Collins, Colorado +1-520-***-**** (available on WhatsApp) ***************@*****.*** LinkedIn: linkedin.com/in/anmol-tripathi-60311917a GitHub: github.com/unit-mole Portfolio: unit-mole.github.io PROFESSIONAL SUMMARY

AI Engineer and Data Scientist with 7+ years of professional experience across network operations, analytics, graduate Data Science, ML research, and applied industrial AI. Currently designs and deploys production LLM, RAG, and Agentic AI systems and multi-stage NLP classifiers for enterprise quality intelligence, built in Python with LangGraph, hybrid retrieval, MCP tool integration, and SQL-based data pipelines. Strong in classical ML and deep learning (PyTorch, LightGBM, LSTM/CNN), evaluation-first engineering, citation and numeric verification, and FastAPI/Docker/CI delivery. M.S. in Data Science, University of Arizona.

TECHNICAL SKILLS

Languages & Data: Python, SQL (SQL Server), pandas, NumPy, DuckDB, Power BI, Tableau Machine Learning & Deep Learning: scikit-learn, LightGBM, PyTorch, Hugging Face Transformers, feature engineering, hyperparameter tuning, model selection, probability calibration, ensembling, classification, time-series (ARIMA/SARIMA), LSTM, Bidirectional LSTM, CNN, Autoencoders

LLMs & Generative AI: LLM fine-tuning (LoRA, QLoRA, PEFT), embeddings (BGE-M3, Sentence Transformers), prompt engineering, context management, LLM evaluation, hallucination mitigation RAG & Agentic AI: LangGraph, stateful multi-step workflows, tool calling, routing, planning, Model Context Protocol

(MCP), document processing and chunking, hybrid retrieval (dense + BM25, RRF), reranking, Qdrant, Neo4j, GraphRAG, human-in-the-loop approval

AI Services & MLOps: FastAPI, REST APIs, Docker, GitHub Actions CI/CD, pytest, Ruff, Bandit, Phoenix tracing, ONNX, model versioning, monitoring, Git

PROFESSIONAL EXPERIENCE

Hach September 2024 – Present

Quality Data Scientist United States

• Designed and deployed an internal Agentic RAG assistant for Quality and R&D users that retrieves source-grounded evidence from historical case records and technical documents. Built incremental ingestion into SQL Server, a Qdrant hybrid index (BGE-M3 + BM25) and a Neo4j temporal graph, orchestrated with LangGraph, an LLM router and read-only MCP tools, with citation and numeric validation.

• Developed and productionized a three-stage NLP classifier on case text plus structured data, combining fine-tuned Transformer representations, word- and character-level sparse features, LightGBM, probability calibration, ensembling and chronological validation. Improved held-out external Defect Symptom accuracy from 69.3% to 70.1%, macro F1 from 56.1% to 57.7%, and high-confidence coverage from 38.4% to 46.0%; scheduled daily scoring writes predictions back to the source database.

• Built a prototype LLM-powered investigation copilot combining statistical analysis, RAG and automated report generation. Fine-tuned with PEFT, LoRA and QLoRA (4-bit NF4) for memory-efficient adapter-based inference; added numerical validation, citation checks, automated tests and human-in-the-loop review, evaluated on accuracy, GPU memory and latency.

• Automated recurring quality analytics across product lines with Python and SQL Server pipelines (SPC, variability and yield analysis, anomaly detection) and delivered Power BI and Excel reporting for Quality, Engineering, R&D and Manufacturing, using Git, Docker, CI/CD and logging for maintainable workflows. The University of Arizona College of Nursing May 2023 – August 2024 Machine Learning Engineer Tucson, Arizona

• Led development of deep-learning and predictive-analytics workflows on longitudinal wearable-ring temperature data from approximately 135 research participants to estimate labor timing; research use only, not clinically validated.

• Designed, trained, tuned and compared SVM, LSTM, Bidirectional LSTM, CNN and Autoencoder models against an existing neural-network baseline, using validation and prediction-error analysis. LSTM sequence modeling performed best and narrowed the estimated labor-prediction window from approximately 14 days to approximately one day in internal evaluations.

• Built the end-to-end workflow: SQL Server extraction, cleaning, temporal feature engineering, sequence generation, ARIMA/SARIMA trend analysis, hyperparameter tuning and architecture comparison, communicating findings and limitations to interdisciplinary collaborators.

Orange Business Services June 2019 – August 2022

Network Operations Center Engineer (progressed from Graduate Engineering Trainee) India

• Developed a predictive machine-learning proof of concept for network-failure detection (approximately 85% accuracy in internal evaluation) and automated recurring data extraction, validation and fault analysis with Python and SQL.

• Analyzed more than 10,000 network-performance metrics daily across Cisco and Juniper environments, performing root-cause investigation and building Tableau dashboards for 5+ critical network KPIs. University of Arizona BookStores September 2022 – August 2023 Student Assistant Manager (promoted from Student Assistant within four months) Tucson, Arizona

• Analyzed 80,000–100,000 monthly sales records across 30+ categories to support inventory and demand planning, contributing to an approximately 15% reduction in overstock and stock-shortage conditions; built Tableau dashboards and mentored 8–10 team members.

SELECTED PROJECTS (PUBLIC PORTFOLIO)

ReliabilityOps – Agentic Root-Cause Investigation LangGraph, Qdrant, BGE-M3 + BM25, RRF, FastAPI, Phoenix

• Open-source agentic RAG system with hybrid retrieval, a claim and citation verifier, human approval step and tracing, served locally with Qwen3-14B. Achieved 93% frozen RCA Top-1 accuracy, 100% citation precision and 12.88s mean latency across 100 test cases.

RepoAtlas – Repository Intelligence and Safe Coding Agent Docker, pytest, Ruff, Bandit, Qwen3-8B

• Built syntax-aware chunking, dependency graphs and test discovery for code retrieval (84.3% File Recall@20); agent edits run in sandboxed Docker containers and are verified with pytest, Ruff and Bandit. FilingsGraph – Temporal Financial GraphRAG DuckDB, Qdrant, NetworkX, question router

• GraphRAG over 20 SEC filings and 121,745 XBRL facts with a question router: 100% router accuracy, 100% fact-selection accuracy and 94.44% textual MRR.

Applied Deep Learning and NLP Portfolio PyTorch, Hugging Face, LoRA/PEFT

• Cross-encoder reranking (MRR@10 +8.1% on SciFact, +10.7% on NFCorpus), schema-aware Text-to-SQL with CodeT5+/LoRA (56.92% execution accuracy on 1,040 held-out queries), and an LSTM autoencoder for industrial failure detection (0.887 ROC-AUC). Code, model cards and demos on GitHub and Hugging Face. EDUCATION

University of Arizona – M.S., Data Science (GPA 3.889/4.0) 2022 – 2023 Texas McCombs School of Business – PGP, Data Science & Business Analytics (GPA 4.00/4.00) 2021 – 2022 Amity University – B.E., Electronics and Telecommunications 2015 – 2019 CERTIFICATIONS

Complete A.I. & Machine Learning, Data Science Bootcamp Google Data Analytics Professional Certificate The Complete SQL Bootcamp Introduction to Python

Contact this candidate