AI Experience

Agentic AI systems I have built.

These projects show how I work across multi-agent orchestration, retrieval-backed systems, live AI participation, and product-grade AI workflows. EasyFlow focuses on tenant-scoped operational agents, IntelliMeet on real-time and post-session AI collaboration, and AskDB on natural-language data access.

EasyFlow

Latest
EasyFlow landing page preview

Open source supply chain coordination platform with visual workflows, isolated workspaces, risk dashboards, automation, and an AI assistant.

IntelliMeet

IntelliMeet platform home page

AI expert simulation platform where custom agents join video sessions, generate transcripts, and power post-session analysis.

Details

AskDB

AskDB

Natural language to SQL query system.

Details

AI Implementation

How AI works inside each project.

Scroll through this section to see how each project uses AI differently: EasyFlow for tenant specific agent orchestration, IntelliMeet for live and post session AI collaboration, and AskDB for schema based natural language querying.

AI Implementation

EasyFlow

EasyFlow uses agentic AI as an operational decision layer. It routes tenant questions into specialist agents, uses MCP tools to retrieve tenant-safe workflow and operational context, then returns grounded recommendations or action paths.

Tenant-Scoped Intake

A user asks a question inside a specific tenant workspace. The system captures tenant identity, module context, active workflow state, and entity focus before any agent runs.

Coordinator Planning

A coordinator agent classifies the request and decides whether the inventory, supplier, logistics, or workflow execution agent should be used.

MCP Tool Retrieval

Specialist agents call MCP-backed tools for inventory risk, shipment events, supplier state, and workflow ownership instead of relying on prompt-only reasoning.

Grounded Response Or Action

The coordinator merges structured findings into one answer. If the user wants action, execution moves to validated backend routes and worker pipelines rather than direct LLM mutation.

AI Implementation

IntelliMeet

IntelliMeet uses AI across two phases: a live expert agent participates during the meeting, then post-session agents process transcripts, build embeddings, and power grounded follow-up Q&A.

Persona Setup

Users define an expert persona through system instructions, which determines how the live AI behaves as an interviewer, architect, mentor, or reviewer.

Realtime Participation

During the session, the AI joins through the OpenAI Realtime API and responds in context using recent turns, meeting metadata, and persona instructions.

Async Transcript Processing

After the meeting, Inngest jobs parse the transcript, structure speaker turns, generate chunks and embeddings, and produce summary insights.

Ask AI Retrieval

When the user asks a follow-up question, the system retrieves the most relevant transcript chunks and sends only those to GPT-4o for grounded answers.

AI Implementation

AskDB

AskDB uses agentic AI as a schema-aware query planner. It interprets natural-language requests, understands the database structure, generates SQL, and returns answers in a form that non-technical users can actually use.

Natural Language Intake

A user asks a plain-English database question such as sales trends, customer counts, or filtered operational records.

Schema Understanding

The AI system inspects schema context so it understands available tables, columns, and relationships before attempting query generation.

SQL Planning And Generation

A planning prompt produces a structured SQL candidate aligned with the database shape rather than guessing from generic language patterns.

Execution And Explanation

The generated query is run, results are returned, and the system explains the output in a user-readable way instead of exposing raw SQL alone.