KAUSTUBH KAR
Kaustubh Kar

I am a Cloud and AI professional currently pursuing an advanced Master of Science in Artificial Intelligence and Machine Learning. My core focus centers on targeting high-scale AI and Machine Learning engineering roles—specializing in orchestrating deterministic multi-agent workflows, fine-tuning large language models, and building automated cloud topologies.

Leveraging a deep technical baseline established across a 9+ year operational trajectory at Amazon, I bridge cloud scaling economics with next-generation agentic solutions.

Review Resume
Kaustubh Kar

// ARCHITECTURE FOCUS

Profile Core

RUNTIME: Python / SQL / AWS SDK
AI FOCUS: Agentic Loops / Fine-Tuning
INFRA: FinOps / S3 / Athena / Lambda

I direct engineering strategies that marry massive enterprise cloud mechanics with precise machine learning automation frameworks. Over my 9+ year tenure at Amazon executing initiatives inside both AWS and Alexa AI ecosystems, I specialize in turning architectural friction into clean operational infrastructure.

My work centers on technical ownership and systemic risk control—ranging from managing cloud escalations for high-value enterprise consumer profiles to optimizing model behaviors via structured Model Context Protocols (MCP) and customized LoRA alignments.

9+ Yrs
Of Industry Experience
1,700+
Hours Automated
97.6%
LLM Precision
Top 800
Of 31,000 Teams

// COMPREHENSIVE TIMELINE OF IMPACT

Technical Customer Success Specialist

Jan 2022 – Sep 2025

Amazon Web Services (AWS)

Responsible for acting as the primary point of contact (POC) and trusted advisor for premium tier cloud accounts. Handled usage monitoring, multi-tier enterprise cost optimization, and rapid security fraud/abuse mitigations. Partnered closely with regional Account Managers, Technical Account Managers (TAMs), and Solutions Architects (SAs) to safeguard complex global workload parameters, verify security alignment, and scale system capacity quotas.

✓ Workflow Automation: Designed and deployed 14 custom automated integration systems and 45 structural regional SOPs, generating an audited savings wall of 1,768 annual operational hours per associate while improving strict SLA compliance via data-driven root cause evaluations.

✓ Spend Optimization: Monitored cloud economics for premium accounts by aggressively driving structural Reserved Instances (RIs) and Savings Plans adoption loops, optimizing predictability metrics across enterprise budgets.

✓ Friction Reduction: Mitigated high-priority escalations across TAM, SA, and engineering pipelines, reducing recurring core platform blockers by 27.5% and fortifying stakeholder trust vectors.

✓ Capabilities Ingestion: Mentored 45 cross-skill associates on production cloud systems best practices, accelerating client enablement parameters and long-term ecosystem value realization.

Machine Learning Data Associate (Alexa AI)

Jul 2017 – Dec 2021

Amazon Development Centre

Administered global algorithm training data pipelines and evaluated active AI machine learning performance signals. Collaborated with core developer and product teams to map raw information telemetry into concrete, measurable enhancements across Alexa's global Natural Language Understanding (NLU) components and query response metrics.

✓ Pipeline Quality: Maintained an individual annotation accuracy floor above 98% for Alexa's baseline NLU streaming pipelines, identifying systematic error trends that reshaped engineering prioritize queues.

✓ Operational Leadership: Led data quality frameworks across 30+ international associates spanning multiple locales, successfully lifting overall operational accuracy metrics from 75.6% to 96.46% and boosting tracking test pass marks from 80% to 95%.

✓ Model Validation: Conduct system evaluations for landmark conversational platform releases (GSR 1.0 & 2.0), logging regression blocks and tuning data processing capacities.

✓ Experience Analytics: Applied granular analytics to escalation queues, successfully raising resolution accuracy metrics from 63.6% to 74.2% on Alexa devices and from 72.2% to 96% for Fire TV systems.

Technical Support Associate

Dec 2014 – Sep 2015

Wipro (TalkTalk U.K.)

Provided technical infrastructure support and resolved complex infrastructure routing cases under tight service level conditions. Analyzed macro support ticket data trends to design and deploy proactive problem avoidance structures and self-help resource indexes.

✓ Queue Control: Handled massive technical queues, successfully debugging 50–60 high-priority network cases daily with a 95%+ First Call Resolution (FCR) velocity, closing 95% of active tracks inside 24 hours at an audited 97% execution grade.

✓ Revenue Realization: Generated a ~14% growth lift and scaled premium services adaptation parameters by 22% via context-aware solution suggestion, while sustaining overall customer satisfaction scores at 95%+.

✓ Workflow Refinement: Streamlined incident workflow handoffs between teams and mentored peer networks, ensuring clear quantifiable advancements across group SLA compliance trends.

System Deployments

FinOps Platform / LLM

AWS Cloud Security Auditor

Grand Finalist
97.6% Score
Unsloth LoRA

Situation: Production cloud topologies frequently suffer unpatched public exposure across IAM and Security Groups.

Task: Standardize an optimization framework to scan multi-tier infrastructures and output closed JSON fixes instantly.

Action: Fine-tuned a 3B parameter model (Qwen2.5) using Supervised Fine-Tuning (SFT) over curated security audit traces.

Result: Realized a validated precision performance floor of 0.9760, reducing operational MTTR from hours to seconds.

→ Launch Presentation
Serverless AI Ingestion

NL Query Agent — Bedrock Nova

Situation: Non-technical business groups experience heavy query turnaround backlogs when fetching from deep cloud storage repositories.

Task: Build a low-overhead interface to process plain English query text directly against backend data files.

Action: Implemented Strands SDK execution paths backed by Nova Lite; configured sizing check boundaries to route tasks to memory Pandas sets or serverless Amazon Athena SQL pipelines.

Result: Attained isolated, multi-format queries secured via Bedrock Guardrails, maintaining audit logs for under $1/month.

→ Launch Presentation
GenAI Pipeline

GitHub Dev Card Generator

Situation: Open-source developer profiles maintain scattered contribution vectors that lack clear skill telemetry visualization maps.

Task: Construct a serverless pipeline that extracts profile data and compiles it into high-end asset cards.

Action: Deployed a FastAPI service on GCP Cloud Run using Gemini 2.5 Flash; engineered Vertex AI Memory Bank upgrades to guarantee cross-session execution persistence.

Result: Standardized an automated profile generator capable of scaling down to zero overhead instances when idle.

→ Launch Presentation
Spatial Data Analytics

Bakery Location Intelligence Agent

Situation: Retail business owners encounter complex data barriers when cross-checking internal sales registries against external physically localized vectors.

Task: Engineer an optimization assistant to interpret spatial performance patterns using decoupled multi-agent frameworks.

Action: Constructed tool attachments via Google ADK and Gemini 3.1 Pro; established Model Context Protocol (MCP) links across BigQuery datasets and live Google Maps APIs.

Result: Delivered real-time data reports synthesizing neighborhood demographics, pricing arrays, and local routing vectors.

→ Launch Presentation
Voice UI / Lambda Integration

Flight Booker Alexa Skill

Situation: Conversational audio applications hit drop-off errors when processing erratic flight array records from live endpoints.

Task: Stand up a low-latency voice command backend that maps user destination strings under narrow SLA windows.

Action: Configured intent slot handlers via the Alexa Skills Kit backed by AWS Lambda (Python 3.12); integrated SerpAPI Google Flights parameters to route schedule listings.

Result: Sustained an evaluation profile maintaining a 95%+ First Call Resolution velocity across 80+ global target hubs.

→ Launch Presentation
SQL / Analytics Framework

Target Brazil E-Commerce Analytics

Google BigQuery
Relational SQL Mapping

Situation: Target required macro operational insights from ~100k records (2016–2018) to optimize logistics, processing times, and regional parameters across Brazil.

Task: Analyze delivery times, demand seasonality loops, and payment behaviors across 8 interconnected relational datasets.

Action: Developed optimized multi-join SQL queries to compile state-level performance matrices, track ordering frequency distributions, and evaluate EMI installment choices.

Result: Captured a 143% YoY revenue surge (R$3.47M to R$8.45M). Pinpointed that while São Paulo contributes 40.5% of demand, remote states experience 2.6–2.8× higher freight boundaries. Discovered 66%+ of orders concentrate between 1 PM and midnight, with 48.9% traveling as single-installments to focus strategic subsidies.

→ Launch Presentation

// HONORS & AWARDS

Rewards & Recognition

IMPACT Guru Award

Awarded globally at AWS for leading 55 structural process updates and complex risk mitigations.

Global IMPACT Legend

Highest tier tracking designation given for scaling cloud workflow efficiency matrices.

AWSome Builder (3x)

Earned multiple honors for building automated serverless support workflows across cloud pipelines.

→ View Award Document

Pinnacle Superstars

Sustained premium elite performance parameters consistently through fiscal cycles 2022–2024.

→ View Award Document

Most Valuable Player (MVP)

Recognized across AWS organizational groups for stellar engineering escalations control.

→ View Award Document

Extra Mile & Quality Awards

Conferred inside the Alexa AI speech division for maintaining a 98% data annotation precision floor.

→ View Award Document

// TECHNICAL ACCREDITATIONS

AWS Certified AI Practitioner (2026)

Validates engineering expertise in orchestrating serverless foundation models, prompt optimization, and deterministic multi-agent workflows on AWS.

→ Launch Verification Modal

GitHub Copilot Certified (2026)

Validates advanced proficiency in leveraging AI-powered pair programming to optimize code generation, streamline debugging workflows, and accelerate software development lifecycles.

→ Launch Verification Modal

AWS Certified Trainer

Official designation verifying comprehensive mastery of AWS cloud infrastructure topologies combined with elite corporate instruction and engineering team capability scaling.

→ View Credential Document

// EDUCATIONAL QUALIFICATIONS

Master of Science (MS)

Computer Science: Artificial Intelligence and Machine Learning

Woolf University — ECTS Accredited EQF7 Framework via Scaler Neovarsity.

→ Launch Academic Record validation file

Bachelor of Technology (B.Tech)

Electronics & Communication Engineering

West Bengal University of Technology (MAKAUT) | 8.24 Cumulative Graduation GPA.

→ Access B.Tech document registry endpoint

// SECURE INGEST CHANNEL

Establish Communication