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 → /academic.
“refactors in prod, apologizes in pull requests.”
— anonymous commit message, probably
the lore (abridged)
everything true, nothing serious
- • 179-page thesis. 3 papers. still counting.
- • ships on fridays. coping is a strategy.
- • allegedly owns a cat named segfault.
jobs i've conned my way into (real)
linkedin's words, my vibes
aka Graduate Researcher (MSc, Software Engineering) at Ontario Tech University — a university that tolerates my 1am commits.
- 🔸Designed TRAC-RE — a Traceable, Reliable, Auditable, and Contextual framework for automated Requirements Engineering, built from four interconnected phases that take 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.
- 🔸Phase 1 — Context-aware requirement identification: empirical study of sentence-level classification (all-mpnet-base-v2, DeBERTa-v3-base) across frozen / LoRA / full fine-tuning and context windows k=0–3, showing structured local context adds +16 F1 points (0.664 → 0.894) and that 4K domain-aligned samples (F1=0.894) beat 15K mixed (F1=0.883). Supporting paper submitted to RE'26 Main.
- 🔸Phase 3 — Governed LLM parsing: built and benchmarked a multi-model parsing pipeline (GPT-4.1, Claude Opus 4, Meta-70B) producing confidence-scored HLJ artifacts with a versioned tag-governance stack (Harvest → Filter → Cluster → Validate → Whitelist → Audit → Drift); tag precision improved 0.657 → 0.897 at the strictest v2 stage. Supporting paper published at CASCON 2025.
aka AI Engineer at Palomino Systems — a place that asked for a migration, got a whole pipeline.
- 🔸Designed and shipped Laravel Upgrader, a fully autonomous CLI-driven AI agent that migrates legacy Laravel codebases end-to-end — a 10-step pipeline (analysis → planning → transformation → validation → self-healing) processing ~10k LOC/run at under $80/run, with 85–90% automated issue resolution via detect → fix → retry loops.
- 🔸Measured two full migrations by hand first, then automated those exact steps — cutting manual migration effort by ~80% (a baseline-grounded number, not an estimate).
- 🔸Owned the codebase end-to-end and was the primary technical contact with Palomino stakeholders, presenting the agent architecture and translating tradeoffs into prioritized recommendations.
aka Founding Engineer at Mediabridge — a place with ads that build themselves (scary).
- 🔸Co-built and launched a multi-tenant SaaS platform from scratch, owning architecture decisions across frontend, backend, and AWS deployment.
- 🔸Personally built the core systems — a custom role-based access control (RBAC) layer, a dynamic form builder, and an event-driven notification system.
- 🔸Owned the full lifecycle from requirements through production launch and served as primary technical contact for the client.
aka Laravel Developer at Finserve Infotech — a place where i speak fluent laravel.
- 🔸Built and delivered 4 ERP / POS systems in Laravel, covering data models, business logic, and client-facing workflows.
papers i wrote when i should've been sleeping
the thesis is real, the sleep isn't
“Toward Automated Requirements Engineering: Empirical and Architectural Foundations for Structured Parsing and Knowledge Discovery”
aka: teaching LLMs manners, with receipts. supervised by Prof. Sanaa Alwidian.
the one that actually got published 🏆
A modular multi-model LLM pipeline that converts raw stakeholder requirements into High-Level JSON (HLJ) artifacts with confidence-scored fields, paired with a versioned…
the one still cooking on the stove 🍳
Explicit keyword extraction — even at audit-grade precision — hits a structural ceiling of roughly 1.5 canonical keywords per HLJ artifact, with Jaccard agreement below…
the one under review (fingers crossed 🤞)
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 doma…
want the write-up with actual numbers and no emoji? → academic mode
things i built (on purpose)
shipped > perfect, always
want the grown-up version with the full stack + story? → corporate mode
the arcade (please touch the exhibits)
four ways to waste time productively
powered by a cheap, rate-limited LLM. be nice, it's on a budget.
type a roast. the AI will roast you back, harder each round. it does not have feelings, but it will pretend to be hurt and then end you.
type your name. a real (cheap, sleepy) AI decides if you and dvip build a unicorn or a merge conflict.
launch an AI startup
no product. no plan. infinite confidence. click to raise money you will absolutely set on fire.
hit generate to receive a perfectly reasonable, totally achievable requirement from a product manager who definitely understands engineering. 🫠
it's 4:55pm on a friday. the button is right there.
psst: there's an achievement system. yes, on a portfolio. no, i will not explain myself. try the konami code. ↑↑↓↓←→←→ b a
the meme museum (peer-reviewed)
generated live · academically accurate · citation: trust me bro
curator's note: if any exhibit shows a cat instead of a meme, the meme API died and the cat is covering the shift. tip him.
skills, tier-list style
scientifically wrong, vibes accurate
ok but actually, here's the real stack →
wall of random flexes
receipts, flex, fine print
framed somewhere. probably.
“Toward Automated Requirements Engineering: Empirical and Architectural Foundations for St…”
supervisor: Prof. Sanaa Alwidian
resolved: ∞
remaining: ∞ + 1
status: it's fine. this is fine.
- coffee/day: 2.4
- tea/day: 1.1 (emergency)
- commits/coffee: 7
- commits/tea: 0.3
- uptime: mine? or the servers?
- sleep SLA: breached every night
- on-call vibe: contemplative
every number on this page is traceable to an mdx file in /content. that's a flex, right? right?
boss-fight: high-five the cat
clear this level → unlock hire-me card
high-five the cat above to unlock. no exceptions. cat policy.
coded at 3am · deployed at 4am · debugged at 4:07am · cat slapped 0×