STATUS: PRODUCTION · 4+ YRS EXPERIENCE

Amandeep

AI/ML Engineer — LLMs & Agentic Production Pipelines

I build and ship machine learning systems that run in production, not just notebooks — from agentic LLM applications to the OCR, ETL, and inference pipelines underneath them.

amanjyani98@gmail.com LinkedIn GitHub

Gurgaon, India · +91 98109 12788

resume.profile

{
  "role": "AI/ML Engineer",
  "experience_years": 4,
  "core_stack": ["Python", "Agents", "LLMs", "Databricks"],
  "specialties": [
    "Agentic AI systems",
    "LLM applications",
    "NLP & intent recognition",
    "ML pipeline design"
  ],
  "currently": "Optum Global Solutions (UHG)",
  "impact": {
    "transcript_validation_time": "8h → 1h",
    "intent_accuracy": "80% → 88%",
    "auth_success_rate": "85% → 90%"
  }
}

02 · SCHEMATechnical Stack

The tools the pipeline runs on

ML / AI

Agentic AI (LangGraph)LLMs (GPT-4, Gemini)NLPClassificationRegressionModel OptimizationOCR (Tesseract)

Languages

PythonSQLPySpark

Libraries

PandasNumPyScikit-learnTensorFlowOpenAI APILangChainLangGraphTransformers

ML Ops

Model TrainingFeature EngineeringData CleaningModel EvaluationA/B Testing

Data / ETL

DatabricksApache SparkETL Pipeline DesignData Modeling

Cloud

Azure (AKS, Key Vault)DatadogBigQuery

Tools

JupyterVS CodePower BIPostmanGit

SDLC

Agile (Scrum/Rally)Cross-functional CollaborationDocumentation

03 · TRAINProfessional Experience

Optum Global Solutions (UnitedHealth Group) · Gurgaon, India

Four roles, one company — the progression is the trajectory.

AI/ML Engineer

Apr 2026 — Present

Designed and deployed an agentic transcript-analysis system using LangGraph and the OpenAI API, automating end-to-end review and intent-tagging — cutting human effort from 8 hours to 1 hour of validation per batch.
Analyzed conversational bot transcripts using multiple clustering schemes and LLMs to recognize caller intent, boosting intent-recognition accuracy from 80% to 88% and lifting authentication success rate from 85% to 90%.
Partnered with Product and Marketing to visualize transcript analytics and surface authentication-gap and call-driver insights, directly informing roadmap and messaging decisions.
Built and optimized LLM-based applications (GPT-4, Gemini) for text extraction, conversational AI, and automated summarization — engineering prompts for consistent structured JSON output in production.
Built a Gemini model-variant classification system using statistical analysis of token patterns to categorize LLM requests into pricing tiers.

Data Engineer

May 2025 — Apr 2026

Designed feature engineering pipelines transforming raw log data into aggregated metrics across multiple dimensions, applying data cleaning, outlier detection, and normalization.
Developed async parallel ML inference pipelines (Python, pandas, NumPy) chunking documents into token-limit blocks for production-scale LLM processing, with error handling and retry logic.
Created user-behavior analytics models aggregating session data, identifying usage patterns, and building segmentation features for cross-functional BI reporting.
Implemented data cleaning and validation frameworks comparing API-sourced metrics against database records, applying scikit-learn anomaly detection to flag inconsistencies.

Software Engineer

Jul 2023 — May 2025

Trained and deployed an OCR model (Tesseract) with pre-processing pipelines (deskew, binarization, noise removal) and evaluation metrics (precision, recall, F1) for production QA.
Collaborated cross-functionally with Finance, Product, and Engineering to translate business requirements into ML-driven solutions with clear documentation and stakeholder presentations.

Training Development Program (TDP II)

Jul 2022 — Jul 2023

Developed a ChatGPT-powered automation for text and conversational applications, training team members on prompt design and LLM best practices.
Built data cleaning and model evaluation workflows ensuring accuracy and reliability in ML outputs.

04 · DEPLOYKey Projects

Shipped, not just prototyped

2026

Agentic Transcript-Analysis System

LangGraph · OpenAI API

Multi-step agentic pipeline orchestrating tool calls and structured outputs across graph nodes to ingest, classify, and summarize conversational transcripts.

Cut required human review from 8 hours to 1 hour of validation per batch by handing routine analysis to the agent.

2025 — 2026

LLM-Powered Text & Conversational Applications

Production LLM applications using GPT-4 and Gemini for text classification, structured extraction, and conversational workflows, with model evaluation and optimization built in.

Designed a variant-classification ML system categorizing requests into 6 tiers using feature engineering on token counts, context length, and usage patterns.

Nov — Dec 2024

ML Inference API

GPT-3.5 / GPT-4 · Python

Async ML inference service with pandas/NumPy preprocessing, prompt optimization, and model evaluation — achieving a 7× throughput improvement via parallel processing.

2024

OCR Model Training & Deployment

Trained and deployed a Tesseract OCR model with a custom pre-processing pipeline and feature engineering for document layout analysis.

Production model evaluation: F1 > 0.92.

05 · EVALUATEEducation & Certifications

Credentials

Malviya National Institute of Technology

M.Tech, Computer Science · Jaipur, India

Sep 2020 — Jul 2022 · GPA 8.28 / 10

Amity University

B.Tech, Computer Science · Gurgaon, India

Jul 2015 — Jul 2019 · GPA 8.14 / 10

Microsoft Azure Fundamentals PluralSight Jun 2024

Python Essentials for MLOps Duke University, Coursera Feb 2024

Machine Learning with Python IBM, Coursera Nov 2022