Projects

// work

Production GenAI systems I've designed and shipped end-to-end — from agentic platforms and RAG pipelines to voice AI and the LLMOps that delivers them. Most are proprietary; case studies describe the architecture and decisions without internal detail.

Agentic AI · Multi-tenant Platform

Agentic AI Platform for Legal Workflows

A multi-tenant platform where every legal matter has an AI assistant — a supervisor agent delegates to specialist sub-agents and offloads long-running automation to a worker tier.

Supervisor + sub-agents
Strict per-tenant isolation
Queue-backed workers

PythonFastAPIMulti-agent orchestrationAzure Service BusPostgreSQL + pgvector+5

Built at CloudLexView case study →

RAG · Retrieval

GraphRAG over Legal Matters

A graph-aware retrieval pipeline that indexes 10K+ legal documents and answers case questions in roughly two seconds.

10K+ docs indexed
+35% retrieval precision
−40% latency

PythonLiteLLMCosmos DBEmbeddings & vector searchFastAPI+1

Built at CloudLexView case study →

Voice AI · Event-driven Backend

AI Voice CallBot (PSTN)

Autonomous phone agents that place and answer real PSTN calls, deliver case updates, and hold a natural back-and-forth conversation.

400+ calls/mo autonomous
−60% response time
Zero dead-air

Azure Communication ServicesAzure Functions (Python)Azure OpenAI (GPT-4o)Azure Speech (STT/TTS, SSML)Redis+2

Built at CloudLexView case study →

Full-stack · RAG

Chronological Medical Summary Generator

Type a case ID and get a dated, row-per-encounter clinical timeline extracted from scanned medical records — exportable as a branded PDF or Word document.

End-to-end RAG → timeline
User-defined columns
Branded PDF / DOCX

FastAPILangChain (StructuredOutputParser)Azure OpenAI (GPT-4o)Azure AI SearchReact+3

Built at CloudLexView case study →

Conversational AI · Full-stack

Embeddable AI Intake Assistant

An embeddable chat widget that drops onto any law-firm website with one script tag — qualifying new leads and answering existing clients’ case questions.

One-script-tag embed
Multi-tenant
No secret in the browser

TypeScriptAzure Bot Framework v4Direct LineMicrosoft WebChatrestify (Node 20)+4

Built at CloudLexView case study →

Platform · LLMOps · DevOps

LLMOps & Deployment Platform

The delivery backbone for the GenAI products — multi-stage Azure pipelines, the right compute for each workload, fast rollbacks, and cost/latency observability.

−22% cloud cost
Zero-downtime rollouts
99.9% uptime

Azure DevOpsAzure Container AppsContainer App JobsAzure FunctionsDocker+3

Built at CloudLexView case study →

Foundations · PyTorch

Transformer LM from Scratch

A modern decoder-only Transformer built from first principles in PyTorch — so the production GenAI stack is never a black box.

BPE · RoPE · SwiGLU · RMSNorm
Pre-norm decoder
CS336 test contracts

PythonPyTorchNumPy

Personal · self-studyView case study →

// more

Also built

LLMOps · Evals

Conversation Evals & Bot Config

An evaluation harness and per-firm configuration store for the conversational AI products.

PythonLLM evaluationConfig store

Built at CloudLex

Conversational AI

Marketing Chatbot (HubSpot)

A lead-generation chatbot integrated with a CRM to capture and route marketing-site visitors.

PythonLangGraphHubSpotAzure OpenAI

Built at CloudLex

Conversational AI

Smart Intake Chatbot

Automated client-intake assistant that qualifies and structures new enquiries.

PythonAzure FunctionsAzure OpenAI

Built at CloudLex

Full-stack · Automation

Document Generation Service

Template-driven generation of legal documents from structured matter data.

PythonFastAPIReactAzure Container Apps

Built at CloudLex

Voice / Document AI

Speech-to-Text + GPT

A transcription-plus-LLM pipeline that turns call audio into structured, searchable summaries.

PythonAzure SpeechAzure OpenAI

Built at CloudLex

RAG · Indexing

Helpdesk Knowledge Indexer

Indexes helpdesk content into a searchable knowledge base for a support assistant.

PythonAzure FunctionsVector search

Built at CloudLex