available for opportunities · New York, USA

> hello world, i'm

I build

AI/ML engineer building LLM pipelines, multi-agent systems and RAG applications, and deploying them on cloud infrastructure that scales. Backed by 2.5 years in production DevOps and 3 peer-reviewed publications.

Résumé
19+
projects shipped
2.5+ yrs
production cloud
100%
uptime for Nokia
3
peer-reviewed papers
[01]selected work

Things I've built & shipped.

Machine learning, generative AI agents, cloud platforms and security tooling, each one taken from dataset or idea through to something that runs.

#01featured
NLP & Generative AI· Feb 2026 – Mar 2026

Fair Lease AI

Designed and deployed a multimodal AI application that uses Amazon Bedrock Nova Vision to automate property damage detection via image diffing across 100+ inspection categories, replacing manual inspection workflows and reducing assessment time from hours to seconds per property. Built a RAG pipeline integrating Tavily AI for real-time repair cost estimation grounded in current market data, cutting estimation turnaround by over 70%. The entire serverless backend (Lambda, DynamoDB, API Gateway and S3) is provisioned with Terraform, with a CI/CD pipeline for one-command, repeatable deploys.

Amazon Bedrock Nova VisionRAGTavily AIAWS LambdaDynamoDBAPI Gateway+2
#02
Machine Learning & Deep Learning· 2026

DistressNet: Vocal Distress Detection

A CNN on log-mel spectrograms that scores vocal distress, combined with an LLM read of the transcript and a judge that adjudicates the two signals.

PyTorchFastAPILog-mel SpectrogramsCNN+4
#03
AI Agents & Productivity Tools· Aug 2025 – Sep 2025

AI Agent for Ticket Triage & Auto-Assignment

Multi-agent CrewAI system with 5+ specialised roles that automates bug classification and assignment, cutting average triage time by 60%.

CrewAILangChainPythonStreamlit+2
#04
NLP & Generative AI· Dec 2025 – Jan 2026

University Q&A Platform

Real-time AI Q&A platform for university questions that grounds LLM answers in online sources and prior community discussions.

Next.jsSupabaseRAGAI APIs+2
#05
NLP & Generative AI· 2026· team

DocGuard & CareerMatch

Full-stack SaaS for sentence-level AI-generated-text detection and ATS résumé scoring, built on a transformer ensemble and FastAPI.

FastAPINext.jsRoBERTaSentence Embeddings+4
#06
AI Agents & Productivity Tools· Jul 2026

Mindgarden

AI creative-ideation canvas built for the IBM AI Builders Challenge. Four AI personas pitch directions and ideas grow, remix and crystallise on an infinite canvas.

Next.jsTypeScriptReact FlowIBM watsonx+3
#07
NLP & Generative AI

Tax Invoice Reader

Created an AI-based tool to extract and process tax invoice data using AWS Bedrock and Pinecone.

AWS BedrockPineconePythonOCR+1
#08
AI Agents & Productivity Tools· Sep 2025

Research Agent

Multi-agent research pipeline that pulls papers from arXiv, summarises them with a transformer model and reviews their credibility.

PythonAutoGenHugging Face TransformersBART+1
#09
AI Agents & Productivity Tools· Oct 2025

AI News Intelligence Dashboard

Agentic pipeline that fetches the latest AI, DevOps and cybersecurity news, summarises each article and scores its sentiment on a live dashboard.

LangChainHugging FaceStreamlitPlotly+2
// more projects10
[02]tech stack

The tools behind the work.

Across the full pipeline: data wrangling, modelling, agent orchestration, and the infrastructure, observability and security it all runs on.

Amazon BedrockLangChainCrewAIAutoGenRAG PipelinesVector DatabasesPrompt EngineeringFine-tuningPyTorchTensorFlow / KerasScikit-learnCNNsNLPHugging FaceApache Spark (PySpark)Hadoop (HDFS)Pandas / NumPyPostgreSQLMongoDBDynamoDB
SupabaseMatplotlib / Plotly / TableauAWSMicrosoft AzureTerraformKubernetesDockerJenkinsGitHub ActionsGrafana / PrometheusSplunkCloud SecurityJira / GitPythonSQLTypeScript / JavaScriptBashYAMLR
~/stack/genai.py
1# Generative AI & MLOps
2stack = {
3"Amazon Bedrock": "Foundation models, Nova Vision and serverless GenAI apps on AWS",
4"LangChain": "Chains, tools and retrieval for LLM applications",
5"CrewAI": "Role-based multi-agent systems for end-to-end automation",
6"AutoGen": "Conversational orchestration of collaborating agents",
7"RAG Pipelines": "Grounding LLM output in retrieved, current sources",
8"Vector Databases": "Embedding storage and semantic search (e.g. Pinecone)",
9"Prompt Engineering": "Template design with automated evaluation of output quality",
10"Fine-tuning": "Adapting pretrained models to domain-specific tasks",
11}
12
[03]experience

Where I've shipped.

Applied AI in legal tech, LLM research at the Fordham EDM Lab, two TA roles, and 2.5 years running production cloud at Hitachi.

commit 1a4a5ba(HEAD → main)· Aug 2026 – Present

AI Intern

Shane Smith LawCharlotte, NC (Remote)
Internship

Building AI tools for a multi-state personal injury practice, finding the manual, repetitive parts of legal work and automating them so attorneys and case managers can focus on clients.

  • +Developing LLM-powered tools that automate document review and summarization of case files and medical records
  • +Building AI-assisted workflows for client intake and case management, streamlining repetitive tasks for attorneys and case managers
  • +Collaborating with the team to design process improvements that reduce turnaround time and improve consistency across daily operations
LLMsDocument AISummarizationPrompt EngineeringWorkflow AutomationPython
commit e6cbb5d· Jan 2026 – Present

Graduate Research Assistant, AI & Automation

Fordham University · EDM LabNew York, NY
Research

Building LLM pipelines that turn raw lecture material into structured course content, with the testing and validation needed to trust the output.

  • +Built an end-to-end LLM pipeline that ingests 100+ PDF lecture documents and auto-generates structured lessons, summaries and quiz questions, cutting manual preparation time by 80%
  • +Engineered prompt templates and evaluated output quality with automated testing frameworks, iterating on prompt design for accuracy, coherence and coverage
  • +Designed CI/CD-inspired ingestion workflows with output validation and coverage-based quality gates across document types and course domains
  • +Containerized the pipeline with Docker for environment parity and faster onboarding of new contributors
LLMsPythonPrompt EngineeringAutomated TestingDockerCI/CD
commit 04d4de6· Aug 2026 – Present

Graduate Teaching Assistant, Data Mining

Fordham UniversityNew York, NY
Teaching

Supporting instruction in core data mining techniques (classification, clustering, association rule mining and model evaluation) for graduate students.

  • +Guide students through hands-on Python assignments and projects covering preprocessing, feature engineering and predictive modeling on real-world datasets
  • +Hold weekly office hours and grade assignments, reinforcing core concepts through one-on-one instruction
  • +Support instructors with lecture preparation, assignment design and grading rubrics for consistent, fair evaluation
PythonScikit-learnPandasFeature EngineeringModel Evaluation
commit 15e989b· Aug 2025 – Jan 2026

Teaching Assistant, Database Management Systems

Fordham UniversityNew York, NY
Teaching

Supported instruction in relational modeling, SQL query optimization, indexing strategies and transaction management for undergraduate and graduate students.

  • +Held weekly office hours and graded assignments for 50+ students covering SQL optimization, indexing, schema design and transactions
  • +Mentored students through semester-long projects in schema design, query tuning and data modeling, with feedback that improved project outcomes
SQLPostgreSQLMySQLER ModelingIndexing
commit 667de10· Jan 2023 – Jul 2025

Cloud Infrastructure Specialist

Hitachi Digital ServicesBangalore, India (Remote)
Industry

Two and a half years running multi-cloud production infrastructure for enterprise clients including Nokia, as release manager for systems serving millions of users.

  • +Served as release manager for Nokia's platform handling millions of concurrent users, sustaining 100% production uptime over 2.5 years with zero critical incidents
  • +Architected multi-cloud (AWS/Azure) environments with Terraform IaC, cutting provisioning time by over 90%, from several hours to under 10 minutes
  • +Managed Kubernetes-orchestrated microservices that provided reliable model-serving infrastructure for data-intensive workloads
  • +Designed and maintained Jenkins CI/CD pipelines across multiple teams, reducing regression cycle time by 40%
  • +Implemented centralized observability with Grafana and Splunk, enabling data-driven incident response and lower MTTR
  • +Enforced cloud security best practices across IAM policies, VPCs and security groups, passing all compliance audits with zero critical findings
AWSAzureTerraformKubernetesDockerJenkinsGitHub ActionsGrafanaSplunkPrometheusIAM
[04]about

The human behind the models.

Sarthak Pokhrel
subject
Sarthak Pokhrel
status
online

I like the whole lifecycle: the messy dataset, the model that finally converges, and the pipeline that ships it without anyone getting paged.

I'm pursuing my M.S. in Data Science at Fordham University, building on a B.Tech in Computer Science (AI) from Jain University. Before grad school I spent 2.5 years as a Cloud Infrastructure Specialist at Hitachi, keeping Nokia's production systems at 100% uptime, which is where I learned to care as much about how models run as how they score.

Today I build LLM tools as an AI Intern at Shane Smith Law, research automated LLM pipelines at Fordham's EDM Lab, and TA graduate Data Mining.

profile.yaml
education:
M.S. Data Science, Fordham University (GPA 3.8/4.0)
undergrad:
B.Tech Computer Science, AI · Jain University (8.6/10)
research:
3 peer-reviewed papers (IJCRT, JASC, ICMLBDA)
focus:
LLMs · multi-agent systems · RAG · MLOps
location:
New York, USA
off_duty:
Piano, hiking
// research, awards & certifications
3 peer-reviewed publications on CNN and NLP diagnostic models
Centennial Scholarship, Fordham University
Study in India Government Scholarship, Jain University
SIEM & SOC Fundamentals
AWS Generative AI Developer, Professional (in progress)
[05]contact

Let's build something intelligent.

Recruiter, collaborator or just curious about the work? I'm always excited to talk AI, data science, SRE or DevOps.

send_message.sh

> awaiting input_