GitHub - Devarsh05/Devarsh05 · GitHub
Skip to content

Latest commit

 

History

7 Commits

Folders and files

Repository files navigation

Devarsh Prajapati

AI / Full-Stack Engineer · Computer Science @ York University · Toronto, ON 🇨🇦

I ship production agentic systems — LLM pipelines, RAG platforms, and real products that measure their own impact.
Currently building SixRise, an AI-search visibility engine for Shopify merchants.

Open to New-Grad / AI Engineer roles — 2026.


AI answer engines (ChatGPT, Perplexity, Gemini) are becoming a primary product-discovery channel — but merchants are blind to whether these assistants recommend them or a competitor, and why. SixRise is a live Shopify embedded app that diagnoses the gap, ships grounded fixes, and proves the uplift.

Founding engineer · solo build · live & deployed in production.

  • 🧠 Designed a graph-shaped async agent pipeline from scratch — eight typed nodes (interrogate → engine-run → extract → share-of-model → audit → optimize → publish → verify) in Python / FastAPI / Arq. No orchestration framework — hand-composed for full control over state, retries, and failure isolation.
  • 🔒 Built a safe write-back system with a hard grounding invariant — the optimizer never fabricates a product attribute, every fix carries a before/after diff + source citation, and publishing goes through a human approval gate before touching a live store.
  • 📐 Engineered a measurement-sacred verifier — the publish timestamp is the immutable anchor for every before/after comparison and is never mutated by the measurement layer, keeping share-of-model uplift honest and reproducible.
  • 🧩 Shipped the Shopify-facing shell in TypeScript / React Router 7 — OAuth, session storage, GDPR compliance webhooks, embedded Polaris UI — talking to the Python brain over an internal authenticated API.
  • ☁️ Deployed to production on Northflank + Neon (Postgres + pgvector) + Upstash Redis; live engine access via Perplexity Sonar and OpenAI structured outputs.

Python FastAPI Arq TypeScript React Router 7 AI Agents LLM PostgreSQL pgvector Redis Shopify OAuth


🚀 More Projects

Understand any codebase in minutes. Paste a GitHub URL and chat with the repo via RAG — with real file/line citations and a dependency-graph visualization.

  • Architected a full-stack RAG system indexing GitHub repos via Tree-sitter AST parsing (function/class-level chunking) across Python, JS & TS — cited, hallucination-resistant Q&A over codebases up to 10,000 files
  • Designed a provider-swappable AI layer (Embedder / LLMClient interfaces) — local open-source models in dev, OpenAI in prod, via config alone
  • Built a production cost-control system — Cloudflare Turnstile + atomic Redis dual rate limiting — powering a public zero-signup demo with bounded API spend
  • Deployed a 5-service backend (FastAPI, Celery, ChromaDB, PostgreSQL, Redis) on Railway with SSE progress streaming + a Next.js frontend, covered by 230+ tests

FastAPI Celery RAG Tree-sitter ChromaDB PostgreSQL Redis Next.js TypeScript Docker

📂 More builds

📈 AI Stock Insights Platform — End-to-end ML pipeline for real-time stock analysis with an interactive Streamlit dashboard. Improved processing efficiency 40% and cut manual data collection 90% via Yahoo Finance API automation. Python Machine Learning Streamlit

🎬 RateFlix — Full-stack desktop app for movie discovery with personalized watchlists and real-time TMDB sync. Optimized a MySQL schema to cut retrieval latency 30%. Java Swing MySQL TMDB API


💼 Experience

AI/ML Intern — Automators Lab · May 2025 – Sep 2025

  • Built multi-agent AI systems with LangGraph & LangChain, integrating GPT, Claude, Gemini & LLaMA
  • Developed 10+ custom Model Context Protocol (MCP) servers connecting agents to APIs, databases & business apps
  • Engineered RAG pipelines for context-aware assistants and document-intelligence workflows

🛠️ Tech Stack

Category Technologies
Languages Python · TypeScript · JavaScript · Java · SQL · R · HTML/CSS
AI / Agents AI Agents · LLM Orchestration · RAG · Grounding & Verification · MCP · LangGraph · LangChain
Backend FastAPI · Node.js · Arq · Celery · REST APIs · Microservices
Frontend React · React Router 7 · Next.js · Streamlit
Data PostgreSQL · pgvector · MySQL · Redis · ChromaDB
Cloud & Infra Northflank · Neon · Upstash · Railway · Vercel · Docker · CI/CD · Git

📊 Impact Highlights

Metric Result
Agent pipeline nodes shipped to production (SixRise) 8 typed stages
Codebase scale indexed (ContextCode) up to 10,000 files
Automated test coverage (ContextCode) 230+ tests
Custom MCP servers built (Automators Lab) 10+
Data pipeline efficiency 40%
Manual data collection eliminated 90%
DB retrieval latency reduction 30%

🌱 Currently

  • Shipping SixRise end-to-end as sole founding engineer
  • CS Hub @ York — full-stack & cloud workshops, hackathons, peer collaboration

Build things that matter. Measure the impact. Iterate.

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors