themeAcademicCorporateTerminalMemeโœ•
โš ๏ธ unserious zone

hi, i'm Dvip.

i train language models and i train my sleep schedule. one of them is vastly outperforming the other. this is a portfolio, but like, the funny version. the real one is up the hall โ†’ /academic.

๐Ÿ“ Oshawa, Ontario, Canada๐ŸŽ“ MSc ยท ontario tech๐Ÿฑ runs on ramen + meows
๐ŸŽฏ daily roast

โ€œrefactors in prod, apologizes in pull requests.โ€

โ€” anonymous commit message, probably

โ˜… hot take
๐Ÿ“„
182
pages of thesis, 0 of self-care
๐Ÿง 
3
papers written. 1 is the cat's.
๐Ÿ“
1,997
json artifacts, 0 therapists
โ˜•
โˆž
coffees. i am the coffee now.
๐Ÿš€ ships models like pizzasโœบ๐Ÿง  NLP gremlinโœบ๐Ÿฑ certified cat whispererโœบ๐Ÿ›  builds things that work at 4amโœบ๐Ÿ“ writes papers โ†’ writes commit messagesโœบ๐Ÿ”ฅ will debug for snacksโœบโœจ 0 bugs (in this tab only)โœบ๐Ÿš€ ships models like pizzasโœบ๐Ÿง  NLP gremlinโœบ๐Ÿฑ certified cat whispererโœบ๐Ÿ›  builds things that work at 4amโœบ๐Ÿ“ writes papers โ†’ writes commit messagesโœบ๐Ÿ”ฅ will debug for snacksโœบโœจ 0 bugs (in this tab only)โœบ
๐Ÿ“–

the lore (abridged)

everything true, nothing serious

๐Ÿงƒ what i actually do
title on paper: AI Engineer โ€” NLP & Intelligent Systems. in practice: i glue transformers to graphs to LLMs to regexes, and then apologize to all of them in the audit log.
๐ŸŽ“ current obsession
MSc thesis at ontario tech, advised by Prof. Sanaa Alwidian. working title is scary-long but the tl;dr is: make LLMs trustworthy for requirements engineering. governance, provenance, audit trails. spicy stuff.
๐Ÿ— day job
migration engineer @ palomino. i shove legacy laravel codebases into a 10-step AI pipeline and feed the output back to the models until they apologize. 85โ€“90% resolution rate. my apology rate: also that.
๐Ÿค coauthors & co-conspirators
one supervisor. one codebase. one cat. that's the team.
๐Ÿ’ค what i'm not
a morning person. a frontend purist. someone who remembers to update his LinkedIn headline. working on one zero of these.
๐Ÿงฉ fun snacks
  • โ€ข 182-page thesis. 3 papers. still counting.
  • โ€ข ships on fridays. coping is a strategy.
  • โ€ข allegedly owns a cat named segfault.
๐Ÿ’ผ

jobs i've conned my way into (real)

linkedin's words, my vibes

research goblin ๐Ÿ”ฌ (lives on the 3rd floor)Sep 2024 โ†’ Present

aka Graduate Researcher (MSc, Software Engineering) at Ontario Tech University โ€” a university that tolerates my 1am commits.

  • ๐Ÿ”ธDesigning a four-study thesis pipeline that automates Requirements Engineering from raw stakeholder prose to implicit domain-knowledge discovery, grounded in an industrial FinTech/SaaS corpus of 110 requirements and 1,997 High-Level JSON (HLJ) artifacts.
  • ๐Ÿ”ธBuilt and benchmarked a multi-model LLM parsing pipeline (GPT-4.1, Claude Opus 4, Meta-70B) with a versioned tag-governance stack (Harvest โ†’ Filter โ†’ Cluster โ†’ Validate โ†’ Whitelist โ†’ Audit โ†’ Drift); Opus 4 and Meta-70B reach F1=0.85 / precision=0.95 at the strictest v2 stage โ€” published at CASCON 2025.
  • ๐Ÿ”ธConducted an empirical study of sentence-level requirement classification (all-mpnet-base-v2, DeBERTa-v3-base) across frozen / LoRA / full fine-tuning and context windows k=0โ€“3, showing structured local context adds +16 F1 points and that 4K domain-aligned samples (F1=0.894) beat 15K mixed (F1=0.883) โ€” submitted to RE'26 Main.
๐Ÿ“ Oshawa, ON ยท ๐Ÿ”ฌ research role
migration wizard ๐Ÿง™ (the legacy code is scared)Nov 2025 โ†’ Present

aka Migration Engineer at Palomino Systems โ€” a place that asked for a migration, got a whole pipeline.

  • ๐Ÿ”ธDesigned and shipped Laravel Upgrader, a fully autonomous CLI-driven AI migration system built on a 10-step pipeline (analysis โ†’ planning โ†’ transformation โ†’ validation โ†’ self-healing) processing 10k+ LOC/run.
  • ๐Ÿ”ธAchieved end-to-end autonomous migration of small-to-medium Laravel applications at under $80 cost/run, with 85โ€“90% automated issue resolution via detect โ†’ fix โ†’ retry loops.
  • ๐Ÿ”ธReduced migration effort by 80%+ and runtime by 40โ€“60%; validation layers cut post-migration defects by 70%+.
๐Ÿ“ Remote
full-stack gremlin ๐Ÿ›  (reports to a canvas)Apr 2025 โ†’ Present

aka Software Engineer at Mediabridge โ€” a place with ads that build themselves (scary).

  • ๐Ÿ”ธLed development of a modular Ad Builder Canvas Engine and dynamic campaign system, reducing frontend effort by 60%+; served as primary technical point of contact for client-side feature discussions.
  • ๐Ÿ”ธEngineered multi-tenant backend services with RBAC access control and event-driven notification delivery; reduced deployment time by 70% via CI/CD automation.
  • ๐Ÿ”ธProposed and prioritized feature roadmap improvements directly with client stakeholders, translating business needs into system decisions.
๐Ÿ“ Remote
laravel samurai ๐Ÿ—ก (slays ERPs by the 4)Jan 2024 โ†’ Dec 2024

aka Laravel Developer at Finserve Infotech โ€” a place where i speak fluent laravel.

  • ๐Ÿ”ธDelivered 4 ERP/POS systems, improving workflow efficiency by 30%+.
๐Ÿ“ India
๐Ÿง 

papers i wrote when i should've been sleeping

the thesis is real, the sleep isn't

๐Ÿ“š thesis-in-progressMSc, Software Engineering ยท Ontario Tech University

โ€œToward Automated Requirements Engineering: Empirical and Architectural Foundations for Structured Parsing and Knowledge Discoveryโ€

aka: teaching LLMs manners, with receipts. supervised by Prof. Sanaa Alwidian.

pages
182
studies
4
papers
3
HLJ bits
1,997
reqs
110
โœ…Improving Reliability of LLMs in RE with Structured Confidence & Tag Governance
CASCON 2025 ยท 2025

the one that actually got published ๐Ÿ†

A modular multi-model LLM pipeline that converts raw stakeholder requirements into High-Level JSON (HLJ) artifacts with confidence-scored fields, paired with a versionedโ€ฆ

#LLM-governance#Requirements-Engineering#Tag-validation
๐ŸณFrom Explicit to Implicit: Towards Traceable Keyword Discovery in Requirements Engineering
RE'26 ยท RE@Next! ยท 2026

the one still cooking on the stove ๐Ÿณ

Explicit keyword extraction โ€” even at audit-grade precision โ€” hits a structural ceiling of roughly 1.5 canonical keywords per HLJ artifact, with Jaccard agreement belowโ€ฆ

#Implicit-knowledge-discovery#Graph-based-NLP#UMAP-/-HDBSCAN
โณTowards Improving Sentence-Level Requirements Identification via Explicit Local Context Modeling
RE'26 ยท Main Track ยท 2026

the one under review (fingers crossed ๐Ÿคž)

An empirical study of sentence-level requirement classification on 110 real-world FinTech and SaaS documents (~5,700 candidate sentences, balanced to 15K mixed / 4K domaโ€ฆ

#Requirements-classification#Local-context-modeling#LoRA-fine-tuning

want the write-up with actual numbers and no emoji? โ†’ academic mode

๐Ÿ› 

things i built (on purpose)

shipped > perfect, always

$ deploy.sh
โ–ธ Autonomous 10-step AI migration pipeline for Laravel codebases.
โ€œtells legacy laravel 'we need to talk' โ€” runs 10 steps of therapy, then migrates. under $80/session. affordable.โ€
stack: Python ยท LLM Orchestration ยท Laravel ยท CLI
Laravel Upgrader
$ run research_pipeline.sh
โ–ธ Four-study pipeline for automated Requirements Engineering โ€” structured parsing, context-aware classification, and implicit keyword discovery.
โ€œmy thesis but as a pipeline. 4 studies, 3 papers, 1,997 json artifacts, 0 sleep. all logged, all audited, all slightly unhinged.โ€
stack: Python ยท PyTorch ยท HuggingFace ยท SBERT / MPNet ยท DeBERTa-v3
RE NLP System โ€” Thesis Pipeline
๐ŸŽฎ

skills, tier-list style

scientifically wrong, vibes accurate

S
python ๐Ÿpytorch ๐Ÿ”ฅcaffeine โ˜•vibes ๐ŸŒˆ
A
typescripttailwindsbertgraph stuffregex (when lucky)
B
php/laraveldockerawssleep schedule (rip)
C
css battlesblog writingclosing 300 tabs
F
waking up before 10hating cats

ok but actually, here's the real stack โ†’

Research Areas
Requirements EngineeringNLP for Software EngineeringLLM GovernanceKnowledge DiscoveryEmpirical SE
Languages
PythonTypeScript / JavaScriptPHP (Laravel)SQLBash
AI / ML
PyTorchHuggingFace TransformersSBERT / MPNetDeBERTa-v3FAISSLoRA Fine-tuningLLM ApplicationsRAG PipelinesUMAP + HDBSCAN
Systems
Distributed PipelinesPipeline OrchestrationFeedback LoopsConstraint-based ValidationAgentic Systems
Backend / Cloud
LaravelFlaskREST APIsMicroservicesRBACDockerAWSCI/CDGitHub Actions
๐Ÿงƒ

wall of random flexes

receipts, flex, fine print

๐Ÿ… AWS Certified Cloud Practitioner
Amazon Web Services ยท 2024

framed somewhere. probably.

๐Ÿ… OVIN Microcredential
Ontario Vehicle Innovation Network ยท 2024

framed somewhere. probably.

๐ŸŽ“ MASc, Software Engineering (Thesis)
Ontario Tech University
2024 โ€“ Present ยท Oshawa, ON, Canada

โ€œToward Automated Requirements Engineering: Empirical and Architectural Foundations for Stโ€ฆโ€

supervisor: Prof. Sanaa Alwidian

๐Ÿ› bug count

resolved: โˆž
remaining: โˆž + 1

status: it's fine. this is fine.

โ˜• fuel metrics
  • coffee/day: 2.4
  • tea/day: 1.1 (emergency)
  • commits/coffee: 7
  • commits/tea: 0.3
๐Ÿ”ฎ konami fact
press โ†‘ โ†‘ โ†“ โ†“ โ† โ†’ โ† โ†’ b a anywhere on this page. you won't regret it.
๐Ÿ“‰ production metrics
  • uptime: mine? or the servers?
  • sleep SLA: breached every night
  • on-call vibe: contemplative
๐Ÿงพ rรฉsumรฉ receipts

every number on this page is traceable to an mdx file in /content. that's a flex, right? right?

๐Ÿ™Œ

boss-fight: high-five the cat

clear this level โ†’ unlock hire-me card

Official High-Five Counterโ„ข
000
๐Ÿ”’ hire-me card locked

high-five the cat above to unlock. no exceptions. cat policy.

coded at 3am ยท deployed at 4am ยท debugged at 4:07am ยท cat slapped 0ร—