shreshth@kharbanda~/portfoliomain
// PORTFOLIO · v2026.05
uptime 5y 364d
00
01
02
03
04
05
06
07
SK · 2026

shreshth
kharbanda.

software engineer @ AWS Bedrock. for 2+ years i’ve been the engineer in the room with 50+ enterprise customers — design, integration, point-of-contact.

looking to do that full-time as a forward-deployed engineer or co-founder.

statusopen to founder-track rolesbasedSeattle, WAemailshreshthkharbanda1@gmail.comcell(404) 510-6272résuméview ↗
shreshth@shreshth.dev : ~/portfolio —
zshlogs~
tab complete↑↓ history⌘L clearbooting...
$ cat

story.txt

§ 01origin · 2019—now
2019 → 2026
2019 · seattle

before any of this, a code-day badge.

i’m 16. a friend drags me to a hackathon. we ship a mobile app that connects orphanages with foster parents. it wins Best in Show at Code Day Seattle 2019. months later i place top-15 nationally in mobile app development.

i learn the lesson that will run my life for the next decade: the gap between an idea and a working thing in someone’s hands is much smaller than people think.

march 2020 · the world goes inside

COVID hits. the café down the block has a phone but no website.

the world goes online overnight. my city’s small businesses do not. brick-and-mortar shops that had never typed “domain registrar” into a search bar suddenly need an internet presence to survive.

“i’ll build you a site. free. you keep the customers.”

i start cold-calling. cafés. salons. a yoga studio. nobody knows who i am. i don’t have an agency, a team, or a clue — just a laptop, a design eye, and a tolerance for getting hung up on. one yes is all i need.

summer 2020 · the agency

one yes turns into JoDi Services.

i’m 17, doing 1am customer support calls between math homework. i write copy, ship websites, debug payment flows, learn what “net-30” means. by month six i’ve shipped a dozen sites and learned how to talk like a business owner before i’ve ever had a business card.

then bigger fish bite: COAR, a political-risk consultancy working in conflict-affected regions. a luxury architecture firm designing flagship spaces for Apple, Louis Vuitton, and Balenciaga. i’m a teenager from Seattle running invoices to a firm with Apple as a client. it doesn’t feel real, so i just keep shipping.

0+
projects shipped
0 yrs
running solo
0 → 1
first real business
2021 → 2024 · the apprenticeships

i collect the hard skills on purpose.

i go to UW for computer science, but i don’t stop building. i intern at Chewy (300% load-speed wins on a site serving 20M customers), at Tesla (rebuild the solar configurator; ship a national installer-finder used in 25+ markets), then land at AWS shipping features on frontier-AI safety systems and quietly building an autonomous engineering agent that 25+ engineers across 3 AWS teams now use to ship faster.

i wanted to know what “good” looked like across the spectrum: 10-person startup, 5-person product team, 5,000-person org. i now know.

today · aws bedrock · responsible ai

i’ve been the customer’s engineer at frontier AI.

the role on paper says “software engineer.” the role in practice has been the closest thing AWS has to a forward-deployed engineer. for two-plus years at Bedrock Guardrails i’ve been the engineer in the room with 50+ enterprise customers — day-0 conversations about what Responsible AI even means for their product, design sessions to fit guardrails into their production systems, hands-on integration with their engineers, and ongoing point-of-contact long after launch. customers like a top US investment bank, a leading market-data & financial-information company, a global asset manager, a 100M-user consumer fitness platform, and regulated enterprises.

my most recent launch is a new public API — a guardrail for AI agents. i owned it end-to-end: scoping, design partners, cross-team alignment with science, PMs, security, and a half-dozen adjacent service teams, and the launch. earlier, i co-led the Standard Tier launch with two other engineers + research + product (+30% accuracy, +25% GPU utilization via cross-region inference, 60+ languages), and helped design + build the framework that onboards new foundation models so customers can use guardrails with them on Bedrock — cutting onboarding from17 days down to 1 and enabling 40+ model launches at re:Invent.

on the side i built Contour, an autonomous engineering agent now used by 25+ engineers across 3 AWS teams— ~25 hours of work a week saved per user. it started as a personal itch and turned into shared infrastructure.

but i miss the part where you cold-call a café. the part where there’s no playbook and no headcount, just a customer and a problem and you. i’m looking for that again — as a forward-deployed engineer or technical co-founder.

$ git log

experience.log

§ 02--oneline --graph
a7f2cd1
feat(bedrock): launched a new API — guardrails for AI agentsHEAD → main
aws bedrock · responsible ai·jun 2023 → present·Seattle, WA

I’ve spent two-plus years at Bedrock Guardrails where the role practically functioned like an "FDE" — though my actual title is Software Engineer II. I’ve been the engineer in the room with 50+ enterprise customers — day-0 scoping conversations on what Responsible AI even means for their product, design sessions to fit guardrails into their production systems, hands-on integration support with their engineers, and ongoing point-of-contact for multiple enterprises post-launch. Customers like Fortune-100 financial institutions, a Fortune-50 consumer platform, and other heavily-regulated enterprises.

My most recent launch is a new public API on Bedrock Guardrails — a guardrail for AI agents. I owned it end-to-end: scoping, design partners, cross-team alignment with science / PMs / security / adjacent service teams, and the launch.

Earlier, I co-led the Bedrock Guardrails Standard Tier launch alongside two other engineers, in partnership with research and product — 30%+ accuracy gains, 25% higher GPU utilization via cross-region inference, 60+ languages — unlocking massive improvements for customers. I also drove part of the design + implementation of a model-agnostic integration framework for Bedrock Guardrails — the system that onboards a new foundation model so customers can use guardrails with it on Bedrock — which cut onboarding from 17 days → 1 day and enabled 40+ model launches at re:Invent.

On the side I built Contour, an autonomous engineering agent now used by 25+ engineers across 3 AWS teams — saving ~25 hours of work a week per user. Started as my own itch, turned into shared infrastructure.

pythontypescriptaws bedrockresponsible aiapi designcustomer devdistributed systems
4e09a3b
feat(tesla): rebuild energy configurator + ship installer findertag: tesla-energy
tesla·jan 2023 → jun 2023·Palo Alto, CA

Rebuilt core flows of the Energy Product Configurator — the tool customers use to design and buy solar panels and Powerwalls — in React/Next/Node/C# for 60,000+ weekly users.

Midway through, I owned end-to-end development of a new certified-installer finder web app now live in 25+ markets, used by 20,000+ people a week.

Got feedback I was performing at a mid-level engineer standard six months in. Tesla moved fast, and so did I.

reactnext.jsreact nativenodec#sql
c1f72a0
feat(uw): lead engineer @ interACT Lab
university of washington·apr 2022 → jun 2024·Seattle, WA

Joined as a research assistant; stayed two years and became lead engineer.

Led the international + Spanish-language rollout of Social Media Test Drive (SMTD), an educational platform helping adolescents learn how to use social media safely. Reached classrooms around the world.

Automated an anonymized data pipeline (Java/MongoDB) and built ML analysis (PyTorch) so researchers could see how kids were actually using the platform.

javascriptpytorchawsd3.jsmongodbjava
38bd5e9
feat(chewy): cache strategy → 300% load-speed win
chewy·jun 2022 → aug 2022·Boston, MA

Joined a fast-moving team supporting tens of millions of customers. Worked on the frontend perf stack — caching strategies in React/Next that improved micro-frontend load speeds by 300%.

Also tracked down and patched a backend bug quietly affecting 300+ internal users a week (Node, Jenkins, Docker, Terraform).

reactnext.jsnodedockerterraformjenkins
7b2810c
feat(codelabs): bias-detection model + opposing-news app
codelabs (microsoft mentor)·jun 2020 → aug 2020·remote

Eight weeks. Four high-schoolers. A Microsoft mentor. We shipped an Android app that surfaced opposing news perspectives so users could see both sides of a headline.

I owned both frontend (Java/XML) and a TensorFlow bias-detection model that hit ~80% accuracy. First time I shipped with a team. Learned how to take a backseat when it made the product better.

javaandroidtensorflowpython
b0a1d24
init(jodi-services): cold-call → digital agency → global clientstag: founder
founder · jodi services·apr 2020 → may 2022·Seattle, WA

Started during the early days of COVID, when brick-and-mortar shops scrambled to stay afloat. No team, no experience — just a laptop and an eye for design.

Cold-called local cafés and salons. Built sites at 1am. Picked up momentum and eventually worked with COAR (political-risk consultancy in conflict regions) and a luxury architecture firm designing flagship spaces for Apple, Louis Vuitton, and Balenciaga.

My first startup. I just didn’t know to call it that yet.

reactreact nativenodepythonawsgcp
$ ls

career/launches/

§ 036 highlights · 2 roles
@AWS Bedrock· 2024 → 20253 launches
$ cat guardrails-for-ai-agents.md2025

Guardrails for AI Agents

new public API · most recent ship

A brand-new public API on Bedrock Guardrails — a safety layer specifically for AI agents, so enterprises can deploy them responsibly in production. I owned it end-to-end: scoping, design partners, cross-team alignment with science / PMs / security / adjacent services, and launch.

pythontypescriptapi designagent safetycross-team
0→1
API launched
GA
shipped
agents
the wedge: safe agents in prod
$ cat bedrock-guardrails-standard-tier.md2024

Bedrock Guardrails — Standard Tier

co-led the launch — engineering + research + product

Co-led the Standard Tier launch across Bedrock Guardrails alongside two other engineers, in partnership with research and product — unlocked massive improvements for customers. Per the public launch post: +15% harmful-content filtering recall, +30% recall on prompt-attack detection, +32% recall on denied topics, and 33–49% recall improvement vs. other commonly available solutions. Multilingual support across 14+ commonly used languages with 78%+ recall.

ml inferencecross-regionresearch × productmultilingual
+30%
recall · prompt attacks
+32%
recall · denied topics
14+
languages w/ 78%+ recall
visit aws.amazon.com/blogs/machine-learning/tailor-responsible-ai-with-new-safeguard-tiers-in-amazon-bedrock-guardrails
$ cat model-agnostic-integration-framework.md2024

Model-agnostic integration framework

foundation-model onboarding for Bedrock Guardrails: 17 days → 1

Drove part of the design and implementation of a model-agnostic integration framework for Bedrock Guardrails — the system that onboards a new foundation model so customers can use guardrails with it on Bedrock. Onboarding used to take 17 days; the framework cut it to 1, which unlocked 40+ model launches at re:Invent.

platformframework designre:inventdev velocity
17→1
days to onboard a new model
40+
launches @ re:Invent
@Tesla Energy· 2023 → 20233 launches
$ cat certified-installer-finder.md2023

Certified Installer Finder

national web app — owned end-to-end

Led development of a new web app helping Tesla customers find certified solar / Powerwall installers near them. Owned end-to-end: backend through production launch. Live in 25+ markets, used by 25,000+ people a week.

reactnext.jsnodec#sql
25K+
weekly active users
25+
markets
E2E
owned backend → launch
visit www.tesla.com/support/certified-installers
$ cat energy-product-configurator.md2023

Energy Product Configurator

the tool customers use to design + buy Tesla solar

Rebuilt core flows of the Energy Product Configurator — the customer-facing tool for designing and buying solar panels + Powerwalls — in React / Next / Node / C#. Used by 60,000+ people a week.

reactnext.jsnodec#
60K+
weekly users
core
flows rebuilt
visit www.tesla.com/energy/design
$ cat post-order-fulfillment-backend.md2023

Post-order fulfillment backend

the system that runs after a customer hits buy

Built backend systems that drive the energy post-order fulfillment journey — because buying a solar system is a lot more complex than "add to cart." Also shipped supporting features in the Tesla mobile app via React Native + Node / REST so customers could track their orders end-to-end.

nodec#restreact native
$ ls

projects/

§ 043 side bets · 2019—now
$ cat projects/contour.ts

Contour

autonomous engineering agent · built on the side at AWS
A long-running AI agent I built on the side that operates continuously across dev + ops workflows — planning, implementing, reviewing, and investigating with persistent memory and structured code context. Started as a personal itch, now shared infrastructure used by 25+ engineers across 3 AWS teams.
25+
engineers · 3 AWS teams
~25h
per week saved per user
30→7m
on-call triage
85%
root-cause accuracy
typescriptagentsaws bedrockpersistent memorymcp
$ cat projects/atma.bot

atma.bot

an AI assistant who already knows you — through iMessage
An agentic personal assistant that lives in iMessage. Handles your inbox, calendar, finances. Becomes your knowledge base — text it anything (image, video, link, voice). Suggests actions it can execute when you give permission. Everyone deserves a personal assistant.
typescriptllm agentsimessagevector db
visit atma.bot
$ cat projects/curiosity.c

Curiosity

multi-threaded search engine in C / C++
A from-scratch search engine with dynamic indexing across 10,000+ documents. Multithreading, big-endian handling, and architecture-neutral data marshalling. Cut server response time by 30% via concurrent request handling with Boost and the STL.
cc++boostmultithreadingsystems
$ ls

hackathons/

§ 054 wins · 2019—2023
#001Meta — Best Hack for Social Good

a mentorship app for under-represented students

Built at a UW hackathon. Connected students from under-represented backgrounds with mentors in their target industries who shared their identity. Meta picked us for social good.

UW · Hackathon · 2023
#002Google — Best Use of Cloud

Together — communities for the identities you carry

Identity is part of us; it’s hard to find communities that speak to who we are. Together connects users to affinity communities and the members in them — chat, schedule meetings, find your people. Google picked it for Best Use of Cloud.

UW · Hackathon · 2022
#003NeuralHacks — 3rd place

COVID-19 detection through lung X-rays

Trained a model to detect COVID-19 from lung X-rays during the first months of the pandemic. Built it in high school while my world was, like everyone’s, falling sideways.

NeuralHacks · 2020
#004Code Day Seattle — Best in Show

a mobile app connecting orphanages with foster parents

Six high-schoolers and a weekend. We built an Android app that streamlined the adoption process between orphanages and foster parents. Won Best in Show. Hooked me on shipping forever.

Code Day Seattle · 2019
$ cat

next.md

§ 06forward-deployed · co-founder
// NOW HIRING ME

looking for the next thing.

for two-plus years at Bedrock Guardrails i’ve been the engineer in the room with 50+ enterprise customers — day-0 scoping, designing solutions that fit Responsible AI into their production systems, helping their engineers integrate, and staying point-of-contact long after the launch. it’s the closest thing AWS has to a forward-deployed engineering role, and i’m looking to do it full-time outside the cloud.

  • forward-deployed engineer at a fast-moving AI / agents / infra company — i’ll embed with your top customers and ship the integration their CTO will brag about
  • technical co-founder for a founder with a deep wedge — i’ll build the product, the data flywheel, and the early team
  • founding engineer at a seed-/series-A team where the first ten hires set the company’s taste
  • any role with these properties: real customer contact, ownership end-to-end, distance-to-impact < 1 week, AI-leveraged
📨 let’s talkview résumé ↗linkedin ↗currently in Seattle · open to relocation
$ cat

contact.json

§ 07reply within 24h

direct lines

say hi

just hit me up. no template, no filter. tell me what you’re building, or what you wish someone would build, or just send a meme.

i reply fast.

cursor [0000, 0000]
session · 1s