Rahul Gehlot

Hi, I'm @Rahul,

who ships web appswith Next.js + AI + Polish

Full-StackAI Integrator

B.Tech CSE student at Vellore Institute of Technology, Bhopal with specialization in AI & ML

I build full-stack web apps and machine learning systems.

Sometimes separately, increasingly together. Hisaab Pro is an offline accounting system I built and shipped to real clients. Skillence is a career platform with a custom PyTorch model and a pure NumPy inference layer I wrote to cut backend deployment crashes to near-zero.

I write a lot of tests.

Because silent bugs in business logic are worse than visible crashes.

I build automated test suites for critical paths—especially when dealing with client financial data or medical triage rules. That instinct follows me into everything.

now

What I'm doing right now

LearningLLM evaluation frameworks - Langfuse, RAGAS
PreparingJLPT N5 in July 2026
ReadingCan't Hurt Me by David Goggins
Thinking aboutWhether RAG or fine-tuning is the right call for domain-specific tasks
Available forFull-time roles, remote opportunities, internships, and freelance projects. I respond same day.

Last updated: May 17, 2026

Work

Where I've worked

Freelance Developer

Hisaab Pro clients · Remote

freelance2026
  • Deployed Hisaab Pro to small business clients
  • Iterated across 2+ product versions based on client feedback - added payroll automation and client-side PDF export
  • Supported real-world deployment with offline-first architecture ensuring data integrity on USB drive removal
Node.jsExpress.jsSQLiteJest

Education

Academic background and specialized certifications.

Education

July 2026Vellore Institute of Technology, Bhopal

B.Tech

Computer Science & Engineering (AI-ML Specialisation)

CGPA 8.67 / 10.0
Data StructuresComputer NetworksAlgorithmsDatabase Management SystemsMachine LearningDeep LearningComputer VisionNatural Language ProcessingOperating Systems
July 2021K.S. Lodha Public Sr. Sec. School, Falna

Higher Senior Secondary Education (XIIth)

CGPA 8.84 / 10.0
May 2019St. Paul's Sr. Sec. School, Falna

Higher Secondary Education (Xth)

CGPA 8.90 / 10.0

Certifications

2026

Claude Code 101

Anthropic

2026

Claude 101

Anthropic

2026

AI Fluency for Small Businesses

Anthropic

2024

Blockchain Developer Certification

IBM Career Education Program

2024

Adobe UI & UX - Graphic Design

Ethnus via Codemithra

2023

HTML, CSS & JavaScript for Web Developers

Johns Hopkins University (Coursera)

Projects

What I've built

Hisaab Pro

live

Jan 2026 - Apr 2026

Why it exists

Small businesses still manage accounts in physical registers or pirated Tally copies. I wanted an offline-first alternative that works without internet, doesn't charge a subscription, and handles GST invoicing without a second tool.

The key decision

Used AES-256 encrypted SQLite (not plain SQLite) and Write-Ahead Logging - because client financial data at rest with no encryption is unacceptable, and WAL protects against corruption during USB drive removal in exactly the environments these businesses work in.

Outcome

Shipped to real clients across 2+ product versions. Built 475 Jest tests - not because it was required, but because silently wrong accounting is worse than no accounting at all.

Node.jsExpress.jsSQLiteJavaScriptJest

PulseSense

in-development

May 2026

Offline-first, 100% private health companion - track vitals, manage medical info (conditions, allergies, medications, emergency contacts), and navigate emergencies with a deterministic rule-based triage engine. 17.8K LOC of TypeScript across 127 files, 13 SQLite tables, 10 triage rules evaluating 18 symptom flags, 69 drug interaction pairs, PDF/CSV export, dark mode. All fully offline, no accounts, no data ever leaves the device.

Key call: Used a static deterministic rule engine (not ML) for the triage system. ML introduces false positives and unpredictability in health guidance - a deterministic system is trivially testable (29 test cases), generates no surprises, and can't learn bad patterns. Exactly what you want when someone uses it during a medical event.

React NativeExpoTypeScriptSQLiteZustandJest

Skillence

live

Jul 2025 - Mar 2026

Full-stack career platform with a 3-stage recommendation pipeline mapping 692 skills to 894 occupations (~1ms inference). PyTorch-trained ML models with custom NumPy inference layer for zero-crash deployment. Automated campus placement engine with 4-variable algorithmic scoring. Job market analytics processing 30,000+ postings with ML salary prediction across 25+ countries.

React 19FastAPIPyTorchNumPy

SwapHub

live

Feb 2024 - May 2024

Campus marketplace for students to buy, sell, donate, and rent items. Full-stack with Express REST APIs, MongoDB/Mongoose data persistence, and Passport.js session auth with secure password hashing. Modular React components with product grids, image upload, client-side validation, and category carousels.

ReactNode.jsExpressMongoDB

This portfolio

live

Ongoing

Interactive portfolio with an embedded RAG chatbot - Gemini 1.5 Flash, Supabase pgvector, hybrid search, Langfuse tracing. All free tier.

Next.jsGeminipgvectorLangfuse

FORGE

halt

Sept 2025 - Halt/Ongoing

Voice-first desktop control system for Windows. Wake-word activated (EDITH/FRIDAY) with Whisper STT, 8 desktop actions (terminal, browser, volume, lock, etc.), AI fallback via OpenRouter, multi-engine TTS cascade, and optional speaker verification via SpeechBrain ECAPA-TDNN. 2,006 lines of tests across 15 files. All in Python 3.12.

PythonWhisperSpeechBrainOpenRouter

Image Captioning

archived

Feb 2024 - May 2024

Encoder-decoder model using ResNet50 + LSTM trained on 1,500 Flickr images. Flask web app with OpenCV processing for real-time AI-generated image captions.

PythonFlaskOpenCVTensorFlow

PyRush

archived

Aug 2023 - Nov 2023

2D platformer with a Mario Maker-inspired level editor - users create, save, and play 100+ custom levels boosting engagement by 40%. Advanced enemy AI, camera tracking, and animation systems with 15% faster level loads.

PythonPygame

Decisions

Why I chose what I chose

AES-256 encrypted SQLite over plain SQLite

vs Unencrypted SQLite or application-layer encryption

Financial data at rest without encryption is unacceptable on client hardware. Page-level AES-256 is battle-tested, zero-config, and leaves no custom crypto attack surface.

Pure NumPy inference over PyTorch serving

vs Serve the PyTorch model directly in production

PyTorch's runtime overhead is unnecessary for static inference. A pure NumPy forward pass is faster, crash-resistant, and eliminated production backend crashes.

Automated test suite for Hisaab Pro

vs Manual QA or post-deployment bug fixing

Silent wrong accounting is worse than a crash. The test suite ensures any regression in payroll or ledger logic fails the build before reaching client books.

Offline-first SQLite over cloud sync for PulseSense

vs Firebase / Supabase with real-time sync

Privacy is the core product. Storing structured health data locally via SQLite avoids cloud database dependencies, mandatory accounts, and network latency.

Skills

My Gear

Languages

HindiNative
EnglishProfessional proficiency

Soft Skills

CommunicationLeadershipSystems ThinkingE2E OwnershipBias for ActionInfluence w/o AuthorityDealing with Ambiguity

Tech Stack

Dev

  • React 19
  • Next.js
  • TailwindCSS
  • REST APIs
  • React Native
  • Expo
  • Zustand
  • Node.js
  • Express.js
  • FastAPI
  • SQLite
  • MongoDB
  • MySQL
  • JavaScript
  • TypeScript
  • Python
  • C++

AI / ML

  • PyTorch
  • NumPy
  • scikit-learn
  • Pandas
  • Gemini API
  • Azure AI
  • OpenCode
  • Claude

Tools

  • Git
  • GitHub
  • Jest
  • VS Code
  • AntiGravity

Design

  • Stitch
  • Figma

Infra

  • Vercel
  • Render

Sharing

Thoughts & writing

Contact

Get in touch

Open to internships, freelance projects, and full-time roles after July 2026. If you've got a problem that needs an ML system or a full-stack product, I respond same day.