Available for opportunities

AI
Engineer

Agentic AI Developer · Multi-Agent Workflows · Custom RAG Pipelines · MCP · FastAPI

Building autonomous AI systems, RAG pipelines, and structured LLM workflows from scratch in Python — deliberate framework usage, full architectural control.

7
Projects Shipped
Hybrid
RAG: pgvector + tsvector
~80%
API Calls Saved
3 Gens
Finance Stack Iterations
harsh@dev:~$
$ python -c "import harsh; harsh.introduce()"
Initializing Harsh Bhanushali — AI Engineer...
✓ Agentic AI: Autonomous systems, RAG, LLM Orchestration
✓ AI Stack: Groq API, Gemini API, Supabase + pgvector, DuckDB, LangGraph, MCP
✓ APIs & UI: FastAPI, Streamlit, DuckDuckGo, Tavily
✓ CLI Tools: Rich, Python, SQLite/JSON storage
✓ Backend: ASP.NET (C#), SQL Server, HTML/CSS/JS
$ harsh.currently_learning()
> ["Docker", "Ollama (local models)", "eval frameworks"]
$ harsh.get_availability()
> "Open to AI Engineer roles | Entry-level | Remote-friendly"

About Me

I'm an AI Engineer who builds agentic AI systems in Python — using frameworks deliberately and purposefully, with full understanding of every architectural layer underneath.

My focus is on RAG pipelines, multi-agent workflows, and cost-aware AI design. I also have hands-on production experience building enterprise web applications during my full-stack engineering internship — shipping 3 apps, improving workflow efficiency by 30%, and rebuilding a complete system under live client requirements.

"I use frameworks deliberately — always understanding the layer beneath before I abstract it away."

Agentic AI Systems
Autonomous agents from scratch
Hybrid RAG & Search
Supabase · pgvector · DuckDB
CLI Engineering
Python · Rich · Modular design
Full-Stack Backend
FastAPI · ASP.NET · SQL

Featured Projects

Shipped Python applications spanning agentic AI, retrieval systems, and backend architecture

hArI v2

RAG Document Intelligence · Deployed
RAG Supabase

Rebuilt from a single-store ChromaDB prototype into a hybrid-search, multi-user system — upload PDFs or CSVs, ask questions, get cited answers.

Hybrid retrieval: pgvector similarity + tsvector full-text via custom Supabase RPC
DuckDB SQL engine replacing pandas.exec() for safe multi-CSV queries
Supabase Auth, per-user chat persistence, citation UI + telemetry feedback
Python Supabase pgvector DuckDB

Multi-Agent Research Pipeline

Agentic Research System · Shipped
Agents Groq

6-agent autonomous research pipeline — Supervisor, Search, Scraper, Summarizer, Critic, Synthesizer — with complexity-based model routing and quota-aware fallback between Groq and Gemini.

Groq (llama-3.3-70b) primary · Gemini (gemini-2.0-flash) fallback
DuckDuckGo search + web scraping per query
CLI displays which model handled each agent in real-time
Python Groq API Gemini API Rich CLI

AI Agent Engine

Autonomous AI System · Shipped
RAG LLM

Production-grade autonomous agent — 4-layer pipeline (cache → pattern router → RAG → LLM) routing roughly 80% of queries with 0 LLM calls. Per-session cost roughly $0.0005.

ChromaDB + SentenceTransformer RAG with roughly 30ms latency
Cost-aware token tracking & daily quota enforcement
Planner → Validator → Executor → Responder pipeline
Python Gemini API ChromaDB SentenceTransformer

FinOS

Full-Stack AI Finance Platform · Shipped
FastAPI Groq

Third-generation finance system — dashboard for passive logging plus a conversational agent for active querying, both writing to the same SQLite store.

Roughly 60% of queries answered by pattern matching, 0 LLM calls; complex queries stream via Groq + SSE
Financial health scoring, spending forecasts, budget threshold alerts
bcrypt-hardened auth with session tokens and CLI-based recovery
Python FastAPI SQLModel Groq API

NextSteps

AI Career Planning App · Shipped
FastAPI LLM

AI career planning app that analyzes a resume against a job description or URL — delivers full gap analysis, skill mapping, and personalized roadmap in under 5–8 minutes. Clean separation between parsing, analysis, and output generation with no hardcoded assumptions about resume or job format.

Resume vs JD gap analysis with skill mapping
URL-based JD parsing — no manual copy-paste
Modular pipeline: parse → analyze → output
Python FastAPI Groq API Tavily

DevMind — MCP Server

MCP Tool Server · Shipped
MCP HITL

A local MCP server exposing 6 developer tools — file ops, Python execution, and JSON utilities — with human-in-the-loop confirmation built into any action that writes or runs code.

6 tools: read_file, write_file, list_directory, run_python_snippet, format_json, count_tokens
HITL guardrails on write and run — no side effects without explicit confirmation
Solved subprocess stdin isolation bug that silently breaks naive MCP implementations
Python MCP SDK tiktoken uv

LangGraph Parallel SQL Runner

Multi-Agent SQL System · Shipped
LangGraph HITL

Multi-question parallel SQL execution using LangGraph's Send API for dynamic fan-out across schema analysis and execution nodes, with inline review before any query fires.

Send-based fan-out with operator.add reducer for parallel branches
Pydantic-validated question and result models
HITL review gate before any query executes against the database
LangGraph Groq API SQLite Pydantic

Technical Expertise

Hands-on skills across Python, agentic AI, and backend engineering — built and tested in real projects

Core Specializations

Agentic AI

Autonomous systems built from scratch

• RAG, LLM Orchestration, Tool-Calling
• Groq API, Gemini API, Supabase
• Cost-aware AI design

CLI Engineering

Modular Python CLI systems

• Python, Rich, JSON storage
• OOP, modular architecture
• Session management, auth

Full-Stack Backend

Enterprise web applications

• FastAPI, ASP.NET (C#), SQL Server
• HTML, CSS, Bootstrap, JS
• Dashboards

Technology Stack

Python
Intermediate+
SQL
Proficient
JavaScript
Familiar
Supabase
pgvector / Auth
DuckDB
In-memory SQL
Groq API
LLM Layer
Streamlit
AI UIs
FastAPI
Backend APIs
LangGraph
Agent Graphs
MCP
Tool Servers
Git / GitHub
Version Control

Certifications & Recognition

B.E. Information Technology
GEC Modasa · 2024
GATE 2025 Qualified
Computer Science Engineering (CSE)

Let's Build Something Real

AI Engineer open to full-time roles and internships at AI-first startups — reach out if you're hiring or just want to talk shop.

Open to Opportunities

Looking for entry-level AI Engineer, ML Engineer, or Python Developer roles at AI-first startups.

Get In Touch

Open Source & Collaboration

Interested in contributing to open source projects or technical partnerships.

View GitHub
harshbhanu0709@gmail.com
Gujarat, India · Open to Remote
Available immediately