Available · Freelance·Mumbai × India·2026

I build
AI agents✦
for the
real world.

Not demos. Not prototypes.
Production multi-agent systems — enterprise analytics,
domain-specific AI, and intelligent automation.
I architect, build, and ship.

See the work →Work with me

bhushan@agentlab

zsh

Booting...

Years in AI

60%

Token Reduction

Tables Optimised

AI Products

Profile
Bhushan Joshi
AI Agent Engineer
Mumbai, India
2024 — Present
Enterprise AI · Freelance

My edge is translating complex, messy business domains into reliable multi-agent workflows that actually ship — not just look good in a slide deck.

Capabilities·06

What I
architect.

End-to-end AI system design. From messy business requirements to production-grade, reliable agent workflows. No hand-waving.

01·Orchestration

🧠

Multi-Agent Orchestration

LangGraph pipelines with orchestrators, workers, memory and feedback loops — built for production edge cases.

LangGraph · LiteLLM · Claude API

02·NL-to-SQL

🗣️

Natural Language → SQL

Schema-aware NL-to-SQL with dynamic normalization and self-correcting generation for complex databases.

PostgreSQL · Prompt Engineering · RAG

03·Prompts as Code

⚙️

Prompt Systems Engineering

Modular, testable prompt configs that cut token usage 60%+ while improving accuracy. Treating prompts like a codebase.

Python · Hierarchical Prompts · Domain Config

04·Domains

📊

Domain-Specific AI Products

AI grounded in real domain knowledge — enterprise analytics, financial tech — where generic models always fail.

Enterprise · FinTech · Agentic Workflows

05·Backend

🔌

Agentic Backend Systems

Scalable backends with RBAC, per-user memory, self-healing agents, and full observability from day one.

FastAPI · Vector DBs · Microservices

06·Consulting

🛠️

AI Product Consulting

Brief to production-ready spec — architecture, data strategy, model selection, build planning. For startups and founders.

Freelance · Startups · MVP Strategy

Featured · Production System

Enterprise AI
Agent System

A production multi-agent system that converts natural language questions into complex SQL analytics — covering performance metrics, reach analysis, decile breakdowns, and quarterly proration across a 14-table PostgreSQL schema.

✦60%+ prompt token reduction through hierarchical restructuring

✦Write-Back system — dynamic runtime prompt modification, no redeployment

✦Full normalization layer handling domain-specific edge cases

✦48+ SQL generation rules encoded across multi-domain config files

LangGraphLiteLLMAnthropic ClaudePostgreSQLHybrid RAGPython

ai_pipeline.py

$ python ai_pipeline.py

60%+

Prompt token reduction

48+

SQL rules encoded

Schema tables optimised