AI Product Engineer

Shreyas Pachpute

I build intelligent systemsfor real-world problems.

I don't start with technology.I start with the problem.

STRATUM00 / PROBLEM

SELECTED WORK 01 / 02SHIPPED AT COMMERCIENT SYNC

Commercient AI

AI agents on top of an ERP-to-CRM integration platform. I built the agents and the plumbing that feeds them.

MY ROLE: AI Engineer
COMPANY: Commercient Sync
WHEN: Jan 2024 — Mar 2026

PROBLEM
SOLUTION
SYSTEM
IMPACT

VISIT PRODUCTwww.commercient.com/ai/agents

SELECTED WORK 02 / 02SHIPPED AT ZYLITIX

DTAXO

[ONE-LINE DESCRIPTION OF WHAT DTAXO DOES]

MY ROLE: [MY ROLE]
COMPANY: ZYLITIX
WHEN: [START] — [END]

PROBLEM
SOLUTION
SYSTEM
IMPACT

VISIT PRODUCT[PRODUCT URL]

04 — HOW I BUILD

A loop,not a list.

Start with what is actually broken, not with a tool.

DRAG THE LOOP · HOVER A STATION

05 — TOOLBOX

I don't collect technologies.I collect ways to solve problems.

AI

models, serving, retrieval, tuning

AGENTS

tool use, planning loops, guardrails

AUTOMATION

schedulers, triggers, human-in-the-loop

WORKFLOWS

state machines, queues, retries

APIS

design, integration, contracts

BACKEND

services, auth, background jobs

FRONTEND

interfaces people actually use

DATA

extraction, structuring, storage

CLOUD

deploy, scale, observe

DEV TOOLS

whatever shortens the loop

Click a tool to see what I built with it. The stack is chosen by the problem, then learned as fast as the problem requires.

06 — CAPABILITY

What shouldwe build?

Pick the problems, then choose a channel. I'll figure out the system.

07 — LAB

Small systems,running.

Three working miniatures of the patterns behind the real products: an agent loop that stops when the goal is met, a rule-based extractor you can edit, and a job queue with retries and a failure budget. They run in your browser, right now.

AGENT LOOPplan → act → observe

Customer asks: where is order #4821?

DOC PIPELINEunstructured → fields

8 extraction rules loaded.

WORKFLOW ENGINEqueue · retries · budget

sync-erpQUEUED
score-leadQUEUED
parse-docQUEUED
notify-crmQUEUED
rebuild-viewQUEUED
index-siteQUEUED

0 done · 0 failed after 3 attempts

08 — CREDENTIALS

Proof, on record.

01AWS Certified Machine Learning – Specialty (MLS-C01)Amazon Web Services[YEAR]Validates the ability to design, build, train, tune, deploy and maintain ML solutions on AWS, and to choose and justify the right approach for a business problem.VERIFY ↗
02Amazon ML Summer School 2023Amazon (India)2023Selective, test-gated program taught by Amazon scientists: four weekends of applied ML modules with live Q&A from senior applied scientists.VERIFY ↗

You havea problem.

Let's build the system.

RESUME