PJ Panshul Jindal Get in touch
Software Engineer at Microsoft

Building systems
that think.

I design and ship production AI infrastructure — LLM pipelines, retrieval-augmented generation, and multi-agent architectures that hold up under real traffic. MS in Computer Science from UIUC.

Download résumé View work
Find me
Panshul Jindal
Panshul Jindal
AI/ML Engineer · USA
10M+
users served
10M+
Users served
40%
Cloud cost cut
98.9%
Recognition accuracy
4.0
GPA at UIUC
About

Engineer first, researcher second.

I'm an AI engineer who ships agent systems to production. My work sits where research meets production: agent runtimes, multi-agent orchestration, retrieval-augmented generation and LLM fine-tuning — built to be observable, cost-aware and maintainable.

At Microsoft I built the agent runtime other teams build on, and the queue that lets an agent hand off to its future self so long-running work survives restarts. Before that, at Kwikpic I built photo-delivery infrastructure serving over 10 million users at 98.9% facial-recognition accuracy, and at UIUC's RailTEC I led research on NL-to-SQL interfaces and computer vision for structural diagnostics.

I care about the unglamorous parts — evaluation harnesses, retrieval quality, latency budgets — because that's usually what separates a demo from a product. I care as much about the quickstart as the architecture.

Oh, and I built Friday — my personal AI assistant. You can actually message him on WhatsApp. Fair warning though: he's very loyal, he'll only talk to his boss. Try your luck ↗

Top 30 AI power user

Recognised across Microsoft — company-wide

MS Computer Science

University of Illinois Urbana-Champaign · 4.0 GPA

Focus areas

LLM systems, RAG, agentic workflows, evaluation

Currently

Software Engineer at Microsoft · Copilot Growth

Open to

Collaboration, speaking, and interesting problems

Experience

Where I've worked

Four roles across product engineering and applied research — each one shipped something real.

Oct 2025 — Present

Software Engineer

Microsoft · Copilot Growth
  • Recognised as a Top 30 AI power user across Microsoft — company-wide
  • Lead AI architecture for Iris, one of the world's largest marketing platforms — set the runtime, orchestration patterns and evaluation bar now adopted by 5 teams across 14 services
  • Built an autonomous campaign-launcher that assumes the role of a human campaign owner, coordinating specialised agents (planner, content, validation) over typed contracts to translate raw intent into live deployments
  • Developed a fault-tolerant agent state machine and queueing system — by serialising execution steps, agents yield and resume in fresh instances, so multi-day campaigns survive infrastructure restarts
  • Led the rollout of code memory across our services, recording each agent's reasoning for a change alongside the change itself — so the next agent or engineer inherits why the code looks that way
  • Own the delivery platform the agents send through — 5 microservices handling ~1B emails and push notifications weekly, with idempotent send-path retries, backpressure and probe-based pod recovery
Agent RuntimeC#AzureKubernetesDistributed Systems
Jun 2025 — Sep 2025

Software Engineer

Visiostack
  • Built a RAG pipeline with OpenAI embeddings and ChromaDB converting natural-language queries into structured JSON
  • Fine-tuned a domain-specific agent via SFT on synthetic data generated through OpenAI's batch API
  • Applied chain-of-thought prompting and few-shot exemplars to widen reliable query coverage
OpenAIRAGFine-tuningChromaDBPython
Jun 2024 — May 2025

Research Assistant

RailTEC, UIUC
  • Developed an LLM-powered query interface translating natural language to SQL for railway diagnostics
  • Designed a monitoring framework pairing PySpark pipelines with LLM-based anomaly analysis
  • Implemented modular backend microservices on Azure for real-time infrastructure health monitoring
LLMNL2SQLPySparkAzurePython
Jan 2023 — Dec 2023

Software Engineer

Kwikpic AI Solutions
  • Built an LLM-driven personalization engine for multi-channel notifications, lifting user retention
  • Developed REST APIs serving 10M+ users at 98.9% facial-recognition accuracy
  • Achieved a 40% AWS cost reduction via auto-scaling and serverless Lambda migration
LLMNode.jsAWSMongoDBComputer Vision
Achievements

Highlights

A few things I'm proud of.

Top 30 AI power user

Recognised across Microsoft — company-wide — for putting AI to work in everyday engineering.

Perfect 4.0 GPA

A flawless academic record through an MS in Computer Science at UIUC.

10M+ users served

Auth and photo-delivery APIs at Kwikpic, with facial recognition taken to 98.9% accuracy.

~1B messages a week

The delivery platform I own at Microsoft — 5 microservices, idempotent retries, backpressure.

Projects

Selected work

Things I built to answer a question I couldn't find a good answer to.

01 / FEATURED

TaskPilot — Agentic Workflow Engine

A modular multi-agent system where autonomous agents plan, execute, and validate tasks against each other. Redis-backed task queues, a FastAPI control plane, and a real-time shared memory store — designed so agents can fail loudly and recover rather than silently drifting.

FastAPIOpenAIRedisDockerMulti-Agent
plan execute validate shared memory redis queue · typed contracts
02

AutoTest — AI Code Coverage

Uses DeepSeekCoder to read a codebase and synthesize unit tests, then runs an auto-refinement loop against failures. Took a target repo from 40% to 100% coverage.

DeepSeekCI/CDPytest
Private repo
03

Eloquent AI

Full-stack RAG chatbot with Pinecone retrieval, WebSocket token streaming, and JWT-backed sessions — built to keep first-token latency low under concurrency.

FastAPIPineconeOpenAI
View on GitHub
04

StackAI

A vector database exposed as a clean REST API — semantic search, kNN similarity, and Cohere embeddings, containerized and ready to drop into a stack.

FastAPICohereDocker
View on GitHub
Toolkit

Tech stack

What I reach for, roughly in order of how often I reach for it.

LLMs & GenAI

OpenAI APIClaudeLangChainLangGraphRAGFine-tuningMulti-Agent

Languages

PythonGoTypeScriptJavaC++SQL

ML & Embeddings

PyTorchTransformersSentence TransformersOpenCVscikit-learn

Vector & Data

PineconeChromaDBFAISSPostgreSQLMongoDBRedis

Backend

FastAPIFlaskDjangoNode.jsgRPCREST

MLOps & Cloud

AWS BedrockSageMakerAzureDockerMLflowW&B
Education

Academic background

Jan 2024 — May 2025

MS, Computer Science

University of Illinois Urbana-Champaign
  • CGPA 4.0 / 4.0
  • Coursework: Machine Learning for Software Engineering, Artificial Intelligence, User-Centered ML
  • Research at RailTEC — NL-to-SQL systems and computer vision for structural diagnostics
4.0 GPAMachine LearningNLPDistributed Systems
Aug 2019 — May 2023

B.Tech, Computer Science

Vellore Institute of Technology (VIT), Vellore
  • Technical Lead, Apple Developers Group — led AI workshops and mentorship for 200+ students
  • Built internal tooling and documentation standards to streamline collaboration for new contributors
B.TechComputer ScienceLeadership
Contact

Let's build something.

I'm always up for a conversation about AI systems, an interesting problem, or a role where the hard part is the engineering. Inbox is open.

panshuljindal@gmail.com Connect on LinkedIn
github.com/panshuljindal @JindalPanshul Résumé