Designed and built a deterministic, replayable evaluation platform for large language models across competitive and cooperative game environments. A generalized game engine supporting turn-based games (Chess, Poker, Blackjack, negotiation scenarios) with pluggable rulesets and agent policies.
I design & build ambitious products that feel inevitable once you see them.
Full-stack engineer, founder, and systems thinker. I turn chaotic ideas into crisp products, layering beautiful UI on top of serious infrastructure and business logic.
I'm Harshit, a full-stack software engineer and multi-time founder who treats every product like an operating system for human behavior. My work spans marketplace design, AI-assisted workflows, education systems, and digital communities.
I move easily between architecture diagrams, Figma mocks, and low-level implementation. The through-line: pairing opinionated design with boringly reliable infrastructure.
Right now I'm especially excited about:
Currently seeking: Full-time Software Engineer positions where I can apply my full-stack expertise, product thinking, and passion for building systems that scale.
A compliant marketplace where fans buy & trade fractional "shares" of college athletes, with instant payouts, royalties, and a rules-first architecture that plays nicely with NCAA/NAIA.
A massive enterprise-grade hotel and restaurant reservation system with real-time guest chat, job applications, VR lounge bookings, and comprehensive concierge services.
An apartment leasing platform featuring luxury boutique residences with floor plans, availability tracking, resident portal, and comprehensive leasing marketplace.
A comprehensive UI/UX platform for Parent Teacher Associations where students, principals, and local PTA boards can access and update school information.
A command center for school PTAs that turns email chaos, spreadsheets, and group chats into a single, opinionated source of truth for events, volunteers, and fundraising.
ReverseRec is a full-stack, diversity-aware recommendation platform that analyzes user content patterns and expands exposure beyond algorithmic echo chambers. Hybrid Node.js + Python services deliver real-time personalization powered by custom-trained models and structured data pipelines.
A personal operating system that uses LLMs as a cognitive layer on top of files, notes, and calendars—so people can query their lives instead of hunting through folders.
A platform connecting students with a passion to students with a curious mind, fostering mentorship and knowledge sharing in educational communities.
An intelligent multi-model interface that unifies Gemini, OpenAI, Claude, and Grok into a single conversational system with real-time model selection and automatic routing. Features a smart AI Auto Router that dynamically chooses the optimal model based on task complexity, cost, modality, and latency.
A multi-tenant email orchestration platform with shared inboxes, AI-assisted triage, branded template building, and full admin governance. Built for teams that need enterprise-grade email management without the enterprise complexity.
Web-based terminal that pairs conversational AI with a real execution environment. The assistant reasons about code, edits files, renders diffs, and executes commands (javac, java, etc.) within persistent workspaces. Built with auth, RBAC, file management, and an integrated editor. Production-grade safety constraints.
DOD/MIL-inspired zero trust security command system with interactive threat matrix, operations console, and mission-aligned access controls. Implements verify-explicitly, least-privilege, and assume-breach principles across multi-domain security visualization (Air, Land, Sea, Cyber). Features real-time telemetry, corridor authorization, perimeter lockdown, and encrypted brief distribution workflows.
An immersive cyberpunk WebGL experience exploring next-generation LLM integration — from contextual agents and semantic validation to multimodal perception and autonomous agent swarms. Built as a narrative-driven single-page journey using Three.js and GSAP ScrollTrigger, geometric wireframes metaphorically represent core capabilities: icosahedrons for knowledge retrieval, torus knots for security guardrails, spheres morphing across vision/audio/lidar modalities. The experience culminates in a dramatic agent swarm convergence.
Custom GPTs require deep prompt engineering, system instruction design, and knowledge base curation to create domain-specific assistants that outperform general-purpose models. I build GPTs with carefully crafted instructions, structured knowledge retrieval, and safety constraints—turning raw language models into production-ready tools for specific use cases. This demonstrates expertise in AI orchestration, prompt optimization, and building AI systems that solve real problems rather than just demoing capabilities.
My voice as Harshit — practical, just, and restorative.
A doctrinally faithful LDS assistant for study, teaching, defense, and spiritual growth.
Harshit Garg's cold, seductive strategist for decoding human behavior and emotional dynamics.
- Opinionated > generic. Good products say "this is the way" and then hold your hand through it.
- Latent complexity, obvious surfaces. Advanced systems should feel simple until you need the power.
- Everything is documentation. Naming, layout, commit messages—they all teach the next person how to extend the system.
- No orphan features. If we can't explain what a feature is for, we either cut it or redesign it.