Speaking engagements and public presentations on AI

SigAgent.AI - Tracing Claude Agents
AI Tinkerers
AI Tinkerers logo

SigAgent.AI - Tracing Claude Agents

Andrei Radulescu-Banu •
AI agents have become powerful and reliable, but you need to build them right. With 50% of new agents now built with Claude Code, the ecosystem has opened up for integration with MCP, marketplaces, plugins, and SKILLS.md. What's missing? AI Agent Evaluation for debugging during development, and for guardrails and monetization during deployment. SigAgent.AI specializes in tracing Claude agents using Open Telemetry and step-by-step, hook-level monitoring. We covered how to build Claude Agents with SigAgent.AI traceability, what works and what doesn't with MCP tools, CLAUDE.md, SKILLS.md and MCP-as-code, and how to stand up agent infrastructure for value-based and usage-based pricing.
Event Details Slides LinkedIn Post
How To Integrate Stripe Into Your AI Application
AI Tinkerers
AI Tinkerers logo

How To Integrate Stripe Into Your AI Application

Andrei Radulescu-Banu •
How did we built Stripe billing into DocRouter.AI? The talk covered why Stripe was the right choice for scaling metered AI usage, our SPU (Service Processing Units) credit system inspired by Databricks DBUs, free tier → plans → a-la-carte credits with consumption waterfall, dynamic pricing config 100% in Stripe metadata, async Stripe API calls in FastAPI, and our MongoDB backend schema.
Event Details LinkedIn Post Blog Post
Engineering with Guardrails: Innovating in Regulated Spaces
Startup Boston Week 2025
Startup Boston Week 2025 logo

Engineering with Guardrails: Innovating in Regulated Spaces

Ilsa Webeck, Andrei Radulescu-Banu, Marjan Monfared •

Building fast doesn’t mean ignoring the rules—especially in healthcare, fintech, AI, or consumer hardware. Our panel explores how to embed compliance into the development process without slowing innovation - for CTOs, technical founders, and engineering leaders operating in regulated industries.

Moderated by Rabeeh Majidi.
Event Details Blog Post Recording LinkedIn Post
An AI Backbone for Document Processing
Mindstone Boston
Mindstone Boston logo

An AI Backbone for Document Processing

Andrei Radulescu-Banu •
At the Sundai MIT hackathon, we recently develop a flow that grades 5th grade quizzes against a configured rubric. DocRouter.AI implements the back end, with the teacher-in-the-loop able to correct the evaluations. We reviewed this hack!

A front-end UI specialized for this use case can be quickly developed with tools like Manus, Bolt AI, Cursor, Windsurf. The DocRouter is the a tech stack accelerator, just managing prompts and evaluations, with human-in-the loop, hiding the complexity of LLM workflows.
Recording with Transcript Presentation Slides
AI Document Workflows
AI Tinkerers & CarGURUS
AI Tinkerers & CarGURUS logo AI Tinkerers & CarGURUS logo

AI Document Workflows

Andrei Radulescu-Banu •
DocRouter.AI is a drag-and-drop UI that transforms unstructured docs into ERP data – integrated with MCP clients like Claude Desktop – building web dashboards, dynamically – for example, allowing consulting companies to optimize internal engineering resources allocated to contracted projects.
LinkedIn Post
Smart Document Router
PyData Boston & MODERNA
PyData Boston & MODERNA logo PyData Boston & MODERNA logo

Smart Document Router

Andrei Radulescu-Banu •
The DocRouter.AI is an open source Document Understanding Tool. It performs data extraction, at scale, for intelligent automation in a number of verticals: Supply Chain, Healthcare, Finance.

It is implemented using:
– NextJS, NextAuth, TailwindCSS
– FastAPI, Pydantic
– AWS, MongoDB
– LiteLLM
– OpenAI, Anthropic, Gemini, Groq/DeepSeek

To Andrei's surprise, the Cursor AI editor wrote most of the DocRouter.AI code! We discussed how DocRouter.AI was coded.
Recording PyData Smart Document Router Presentation LinkedIn Post
AI Human-in-the-Loop Panel
AI Furnace & Aethos

AI Human-in-the-Loop Panel

Andrei Radulescu-Banu, Dipali Trivedi, Brian Benedict •
Panel discussion on how human-in-the-loop is implemented in practice, when used during LLM development vs in end products, LLM model evaluation, and the importance of humans understanding AI-generated code. Moderated by Nico van Wijk.
Andrei's Blog Post
LLM Orchestration At Scale
AI Camp
AI Camp logo

LLM Orchestration At Scale

Andrei Radulescu-Banu •
Why do we need to orchestrate Large Language Models (LLMs)? We will talk about Perplexity.AI's approach, shared openly in their podcasts – and at enterprise workflow automation with unstructured data at scale.
Event Details