00
MANAN SHAH — AI ENGINEER
◈
MANAN SHAH · AI ENGINEER @ PAYPAL · SAN FRANCISCO, CA

Building LLM systems that hold up in production.

Three-plus years of software engineering across fintech and enterprise. I build RAG pipelines, agentic workflows and the backends underneath them at PayPal, and ship products of my own that are live on the internet.

Manan Shah on the Brooklyn Bridge, Manhattan skyline behind
BROOKLYN BRIDGE, NYC
01

About

I am an AI engineer with a backend engineer's habits. At PayPal I build the LLM services that help support teams reason about merchant disputes: retrieval over years of resolved cases, grounded generation on top of it, and the unglamorous pipeline work that makes both trustworthy. Before that, three years on enterprise insurance platforms taught me what production actually costs. I prototype fast, then harden the result until it holds. Ambiguous problem in, working system out.

CURRENTLYSoftware Engineer at PayPal, AI and LLM systems
FOCUSRAG, agents, embeddings, semantic search
EDUCATIONMS Computer Science, Illinois Tech, 2025
BASESan Francisco, CA, open to relocate
0+
YEARS SHIPPING PRODUCTION SOFTWARE
0
DOMAINS: FINTECH AND ENTERPRISE INSURANCE
0
PRODUCTS DESIGNED, BUILT AND SHIPPED SOLO
0
LIVE DOMAINS RUNNING IN PUBLIC TODAY
02

Experience

A TIMELINE IS JUST DEPTH, TRAVELED
2025 APR 2025 — PRESENT

Software Engineer AI / LLM PLATFORM

PAYPAL · SAN FRANCISCO, CA
  • Built a Python and FastAPI AI service that helps support teams analyze merchant disputes and retrieve relevant resolutions from previously resolved cases.
  • Created a Retrieval-Augmented Generation pipeline using Amazon Bedrock embeddings and Pinecone for semantic search across historical dispute cases and supporting documents.
  • Developed document processing workflows to clean dispute data, chunk large case records, generate embeddings and store metadata for filtered retrieval.
  • Integrated Amazon Bedrock language models with retrieved case context to generate concise resolution summaries and suggested next steps for support agents.
  • Added prompt templates, structured outputs, retrieval filters and validation checks to keep generated recommendations relevant and grounded in dispute data.
  • Used Apache Kafka to process new dispute events asynchronously and trigger background indexing workflows without slowing down customer-facing APIs.
  • Containerized the application with Docker and deployed it on AWS using Kubernetes, GitHub Actions and CloudWatch for CI/CD, monitoring and production troubleshooting.
PythonFastAPIAmazon BedrockPineconeKafkaKubernetes
2021 MAR 2021 — JUL 2023

Software Engineer ENTERPRISE PLATFORMS

DUCK CREEK TECHNOLOGIES
  • Developed Java, Spring Boot and Hibernate services for policy administration, billing and claims workflows on an enterprise P&C insurance platform.
  • Designed REST APIs for quote, policy, billing and claims workflows, and integrated backend services with payment processors and third-party insurance systems.
  • Worked with PostgreSQL and MongoDB to store policy and transaction data, optimizing queries and indexes for high-volume application workflows.
  • Built asynchronous processing workflows using Apache Kafka to exchange policy and claims events between distributed backend services.
  • Helped break parts of a legacy Java application into independently deployable Spring Boot microservices, improving maintainability and release flexibility.
  • Investigated production issues using application logs, API traces and database queries, identifying root causes across distributed services and third-party integrations.
Java 17Spring BootHibernatePostgreSQLMongoDBKafka
03

Projects

SHIPPED. LIVE. NOT SLIDEWARE.
01
LIVE

TrailMail

A Chrome extension that predicts work emails from public company patterns and sends personalized outreach through Gmail. Entirely client side, send-only OAuth scope, open and click tracking, no server ever holding your data. Published on the Chrome Web Store.

JavaScriptGmail APIHunter.ioChrome MV3OAuth2
02
LIVE

DialDo

Give your Mac a phone number. A local-first voice agent: you call a Twilio number, a cloud gateway routes the intent to an Electron desktop client, and it finds files, summarizes PDFs, writes documents and sends email. Plugin-based tool registry, SQLite task history, signed and notarized for macOS.

TypeScriptElectronTwilioLLM APIsSQLite
03
LIVE

GitPixelPush

Draw pixel art on your GitHub contribution graph. A browser-side engine that writes back dated commits through the Git Data API: no clone, no server, and your token never leaves the page. Served from a Cloudflare Worker that also handles the OAuth exchange.

Git Data APICloudflare WorkersOAuthVanilla JS
04
LIVE

Reporun

A desktop preflight tool for public GitHub and GitLab repositories. It reads project manifests, compares requirements with your Mac or Windows PC, and presents setup commands for review before you run unfamiliar code.

Desktop appGitHubGitLabDeveloper tools
05
OPEN SOURCE

OutreachOS

A multi-step LinkedIn cadence engine in the browser: visit, connect, message and follow, sequenced with delays and daily caps. Runs against the Voyager API rather than clicking the DOM, which is the only approach that survives a redesign. 158 unit tests behind the scheduler.

Chrome MV3Voyager APIService WorkersIndexedDB
06
IN DEVELOPMENT

Relay

A two-lane outreach engine: emails go out up front while the LinkedIn crawl runs in the background, so the slow lane never blocks the fast one. Includes a template studio with attachments and live preview, plus a credit strategy that spends per company instead of per person. 773 tests.

Chrome MV3Gmail APIJob QueuesVitest
PRIVATE BETA
07
ENTERPRISE POC

Agentic Voice System

End-to-end voice agent orchestration built for an enterprise prospect. LLM reasoning, webhooks and CRM-style data handoffs, with conversation logic that survives multi-step tasks without losing the plot. The repeatable agentic pattern underneath it became DialDo.

PythonBland AIn8nWebhooksLLM APIs
CASE STUDY ON REQUEST
08
INTERNAL TOOL

Email Research Engine

The inference backend behind TrailMail. A Flask service that resolves a company name to its real domain, mines public pages and search snippets for example addresses, then infers the organization's address pattern and scores its own confidence. One paid lookup per company, never per person.

PythonFlaskBeautifulSoupSSEHunter.io
POWERS TRAILMAIL
09
RESEARCH BUILD

ConnectPilot

A LinkedIn connection automation extension, and mostly an exercise in reverse engineering: DOM clicking proved to be a dead end, so I built request-capture tooling, found the real invitation endpoint, and drove it directly with the correct payload shape.

Chrome MV3webRequestVoyager API
PRIVATE REPO
10
YOU ARE HERE

This Portfolio

The world you are flying over. A procedural dusk coastline in Three.js with real planar water reflections, merged city geometry, a lathe-turned aircraft and a generative WebAudio score. Zero downloaded assets: every texture is drawn to a canvas at boot.

Three.jsGLSLGSAPWebAudio
LOOK DOWN ↓
11

Your idea
here

The next one gets built at your company. I prototype fast and ship things that survive contact with production.

START A CONVERSATION ↗
04

Stack

TOOLS ARE TEMPORARY. SYSTEMS THINKING ISN’T.
LLMS · RAG · PROMPT ENGINEERING · EMBEDDINGS · SEMANTIC SEARCH · AI AGENTS · TOOL CALLING · MCP · LLM EVALUATION · 
PYTHON · JAVA 17 · TYPESCRIPT · SQL · FASTAPI · SPRING BOOT · REST APIS · MICROSERVICES · APACHE KAFKA · 
AMAZON BEDROCK · PINECONE · LANGCHAIN · LANGGRAPH · AWS EKS · LAMBDA · DOCKER · KUBERNETES · GITHUB ACTIONS · 

PROGRAMMING LANGUAGES

Python, Java 17, TypeScript, JavaScript, SQL.

GENERATIVE AI & LLMS

LLMs, RAG, prompt engineering, embeddings, semantic search, AI agents, tool calling, MCP.

AI FRAMEWORKS & PLATFORMS

LangChain, LangGraph, Amazon Bedrock, OpenAI API, Anthropic Claude, Hugging Face.

DATABASES & VECTOR SEARCH

Pinecone, ChromaDB, PostgreSQL, MongoDB, Redis.

BACKEND & APIS

FastAPI, Spring Boot, REST APIs, microservices, Apache Kafka.

CLOUD & MLOPS

AWS (EC2, S3, RDS, EKS, Lambda, CloudWatch), Docker, Kubernetes, GitHub Actions, CI/CD.

TOOLS & PRACTICES

Git, Postman, JIRA, Agile Scrum, system design, LLM evaluation.

05

Education

THREE DEGREES OF PREPARATION
2025 Illinois Institute of Technology MS, Computer Science · Chicago, IL · Aug 2023 — May 2025
2022 K J Somaiya Institute of Management Master of Computer Applications · Mumbai · May 2020 — Jul 2022
2020 K J Somaiya College of Science and Commerce BS, Information Technology · Mumbai · Jun 2017 — May 2020
1st

Hackathon winner. First place among 200+ teams, Duck Creek Hatchathon 2022.

50+

Outstanding academic coach. Illinois Tech, 50+ students mentored.

AZ

Certified. Microsoft Azure Fundamentals and SQL for Data Science.

THE PART WHERE YOU EMAIL ME

Let’s build something great.

mananshah03691@gmail.com
SAN FRANCISCO, CA · OPEN TO RELOCATE · AI ENGINEER @ PAYPAL · RESPONDS FAST