Data.

Engineer.

Two years shipping .NET and Java services in enterprise legal tech at Knovos — ingestion at 5M+ files, APIs over 2M+ records, an OCR pipeline that retired a $200K/yr vendor. Now a Data & AI Engineering intern at BeOne Medicines, building a SQL lineage system that grounds an LLM in a real dependency graph. On the side I build multi-agent LLM systems like Verdict, an AI courtroom for settling disagreements. Finishing an MS in Data Science at UMBC.

ConnectSee Work

Parth is finishing an MS in Data Science at UMBC with a 4.0 GPA.

Always shipping,
always learning.

Find My Work

V

Verdict

2026

An AI courtroom for everyday disagreements: advocate agents debate, an LLM judge rules with confidence scoring and steelmanned cases.

multi-agentBYOK5

Next.js · TypeScript · FastAPI · Python · Anthropic API · OpenAI API · Zod · Pydantic · SSE

M

MindsetLens

2025

Teacher analytics platform classifying student learning mindsets with a HuggingFace transformer and LangChain orchestration.

82%<200ms200+

Python · FastAPI · React · HuggingFace · LangChain · PostgreSQL · Docker · AWS Lambda

ACRB

AI Code Review Bot

2025

Multi-mode automated PR review on Cloudflare Workers with Durable Objects for per-PR state.

TypeScript · Cloudflare Workers · Durable Objects · SQLite

PES

PII Extraction System

2023

Production legal-doc NLP with rule-based + statistical classifiers, 60+ entity types, Solr-indexed.

94%60+

Java · Stanford NLP · OpenNLP · Apache Solr

TPO

Tuition Payoff Optimizer

2025

Loan-payoff vs investment comparison via Monte Carlo simulation over market-return uncertainty.

TypeScript · React · FastAPI · Python · Docker

See everything on GitHub

Tracing My Work Lineage.

Data & AI Engineering Intern @ BeOne Medicines

Jun 2026 – Present (2 Months)

Building a stored-procedure lineage system over the EU Redshift warehouse.

  • Built an end-to-end pipeline that extracts stored procedures from Redshift, parses them with sqlglot, and assembles a directed dependency graph in networkx — 740 nodes, 2540 edges across EU database
  • Designed a deterministic graph layer beneath the LLM assistant so lineage answers resolve against the DAG instead of retrieved text, removing a class of hallucinated dependency claims
  • Generated per-procedure markdown docs with an LLM, published to SharePoint, surfaced through a Copilot Studio agent for non-technical analysts
  • Re-architected delivery from direct-API to the internal Microsoft stack under an API-key constraint without losing the deterministic core

Python · SQL · Redshift · sqlglot · networkx · AWS EC2 · Copilot Studio · SharePoint

Software Engineer @ Knovos

May 2023 – Jul 2025 (~2 Years 2 Months)

Enterprise data ingestion, REST APIs, and ML pipelines for eDiscovery.

  • Enterprise data ingestion in .NET/C# (Slack, Teams, PST, forensic formats) — 5M+ files, 99.5% uptime, 40% faster after a multithreading rework
  • REST APIs for the eDiscovery platform over 2M+ records — 40% throughput up, 35% latency down via Redis caching, query optimization, async processing
  • Replaced a $200K/yr OCR vendor with an in-house Microsoft OCR pipeline — 98% accuracy, 90% faster, 1M+ files/yr
  • Java Spring Boot redaction microservice, 1M+ files/yr with inverse-redaction logic, 20% fewer errors
  • Diagnosed and fixed race conditions in a failing pre-processor two weeks before a court deadline in an unfamiliar codebase — shipped stable, zero crashes at trial
  • Kerberos SSO with Active Directory over C#/LDAP for 200+ users

C# · .NET · Java · Spring Boot · Redis · PostgreSQL · SQL Server · Kerberos/LDAP

Software Engineering Intern @ Knovos

Jan 2023 – Apr 2023 (3 Months)

NLP PII extraction, distributed ETL, and enterprise search.

  • NLP PII extraction with Stanford NLP + OpenNLP, 60+ entity types, 94% accuracy
  • Distributed ETL in Java + Hadoop MapReduce over 100K+ documents, custom InputFormats, 70% faster
  • Apache Solr search with custom analyzers and faceted search — sub-second over 500K+ docs, 97% recall

Java · Hadoop MapReduce · Apache Solr · Stanford NLP · OpenNLP

Some Of My Interesting Stats

5M+files processed
$200Kvendor cost saved
740-nodelineage graph
4.0GPA
100K+docs indexed

Education

UMBC

MS Data Science (CS track)

4.00 GPAAug 2025 – May 2027

Institute of Advanced Research

BTech Information Technology

3.80 GPA2019 – Apr 2023

Get in touch

parthmv1@umbc.edu
GitHubLinkedIn
View Resume

Open to Fall 2026 internships and 2027 new-grad roles