Software engineer in Long Beach

ShreyanshSharma

I build products with UI, backend, data, and ML pieces that have to work together.

Most of my work starts with a simple user problem, then turns into routing logic, database shape, API edges, and testing details. I care about the parts people see and the parts they should never have to think about.

EmailBest way to reach me

LinkedInWork and background

ResumeLatest PDF

GitHubCode and repos

Scroll to explore

Current product

RETAIN AI · Founder & Solo Engineer · 2025–Present

RETAIN AI

Messaging and callback assistant for local service businesses.

RETAIN AI helps local businesses answer missed texts and calls. RETAINchat handles SMS and web messages. RETAINvoice answers calls, places callbacks, checks business rules, responds out loud, logs the transcript, and hands off messy calls to the owner.

Stack

Next.jsTypeScriptTailwind CSSSupabaseGeminiTwilioStripeAWS

Open platform →

RETAINvoice

Live

inbound call · Cedar Dental Studio

inbound support call

0:00 / 0:00appointment_booking

dental · scheduling🇺🇸 en-US[live]ag_cedar01

analyzing waveform…

connecting call…

latency320mst_resp avg

packets0rtp0 · 20ms

resolutionfirstcallbooked

resolved · bookedt_resp avg 320ms

10,000+

Scenario variations analyzed

4M+

Message exchanges stress-tested

150–250ms

p95 response time under load

500+

Concurrent users stress-tested

<0.4%

Error rate under heavy load

Multi-tenant

Runtime-configurable per business

RETAINchat

SMS and web messages use business FAQs, hours, tone, memory, and escalation rules before Gemini writes a reply.

RETAINvoice

Answers inbound calls and callbacks, listens to the caller, runs the same business rules as chat, speaks responses out loud, and flags owner handoff when booking or pricing needs a person.

Owner handoff

Messy conversations are logged with a summary and routed back to the business instead of pretending the AI can finish everything.

Integrations

SupabaseGeminiTwilioStripeVercel

/work

Selected work

Developer tool

Live

scopekit-sandy.vercel.app

Developer Tool

ScopeKit

ScopeKit gives coding agents a small repo pack before they start searching through everything.

I built the CLI to index a repo, choose the useful files, and write a context pack an agent can use immediately.

Numbers

~1,280×

Less context in the auth task

18×

Smaller than Graphify best-effort

No

LLM routing

npm

Published package

Specs

Measured auth-discovery task

8 files / 7 concepts recovered

Codex usage parsed

No LLM routing

Stack

TypeScriptNode.jsCLIMCP ServerZodMCP SDKVercel

Features

01

Picks the files, symbols, tests, risks, and commands for one task.

02

Writes local setup notes for Claude, Codex, Cursor, and MCP.

03

Runs locally. No API key. No LLM routing.

04

Reduces the context agents see before they spend tokens.

Links

Website GitHub npm

Open demo

Aid flow

Live

resilient-aid.vercel.app

Aid protocol

ResilientAid

ResilientAid lets donors fund aid, vendors fulfill it, and beneficiaries receive help without the money disappearing in the middle.

I built the role-based app, wallet connection, vendor redemption path, and transaction trail.

Numbers

425 tx/s

Network throughput

0%

Platform fees

<2s

Settlement time

100%

On-chain transparency

4

Stakeholder interfaces

Live

On Polygon Amoy

Stack

PolygonSmart contractsAid paymentsVendor POSStablecoin settlement

Features

01

Separate screens for admins, donors, beneficiaries, and vendors.

02

Donors can see where funds move instead of trusting a dashboard total.

03

Vendors can redeem credits quickly at checkout.

04

Polygon Amoy records each transfer in the aid flow.

Interfaces

Admin PortalOpen Donor PlatformOpen Beneficiary AppOpen Vendor POSOpen

Open demo

Marketplace demo

Live

artisian.vercel.app/trends

AI marketplace

KalaAI

KalaAI is a marketplace for handmade goods where shoppers can browse products and get an AI guess at the craft origin from a photo.

I built the feed, marketplace pages, trend view, account flow, and image-analysis path as one product.

Numbers

120+

Cultural detection categories

5

Product surfaces

AI

Origin detection from imagery

Live

Deployed on Vercel

Specs

Home

Shorts Feed

Marketplace

Trends

Account

Stack

AIComputer VisionFull-stackMarketplaceNext.jsVercel

Features

01

Shorts-style feed for artisan stories and products.

02

Listings, trends, seller context, and account pages live in one app.

03

Photo analysis maps products to 120+ craft categories.

04

AI results appear inside shopping instead of a separate demo page.

Open demo

Verification flow

Live

veri-degree.vercel.app

Credential verification

VeriDegree

VeriDegree lets a school issue a credential and lets an employer check that it is real without seeing more student data than necessary.

I built the issuer, student, and employer paths around Algorand credential state.

Numbers

Mainnet

Algorand deployment

Soulbound

Academic records

Instant

Employer verification

Private

Selective disclosure

Specs

Algorand credential issuance

Issuer / student / employer flow

Wallet-based credential state

Selective disclosure

Stack

AlgorandSoulbound credentialsZero-knowledge privacyAcademic verification

Features

01

Schools issue records once instead of emailing PDFs again and again.

02

Students keep wallet-based academic records they can present later.

03

Employers check credential status from a public verification page.

04

Private fields stay hidden unless they are needed.

Open demo

/archive

Other technical work

Form scoring demo

Pose to feedback

Coaching output

score

Depth82

Alignment77

Tempo88

Computer vision

AI Motion Analysis

AI Motion Analysis reads pose keypoints from workout video and turns them into form scores a person can understand.

Features

01

Tracks 17 body keypoints across video frames.

02

Scores depth, balance, tempo, and alignment.

03

Turns pose output into short coaching notes.

04

Flags the form issues instead of dumping raw keypoints.

Specs

17-point MoveNet pose sequences

PyTorch scoring pipeline

Readable coaching output

Stack

PythonPyTorchMoveNetOpenCV

View GitHub

Classifier review

5 classes

Sample 1Ready

Sample 2Ready

Sample 3Ready

Sample 4Ready

Model output

eval run

Superficial91%

Parabasal82%

Metaplastic88%

Medical ML

Cervical Cell CNN

Cervical Cell CNN classifies Pap smear images into five cell types and keeps the training path repeatable.

Features

01

Trained on 4,000 labeled images across five classes.

02

Preprocessing stays consistent before training and evaluation.

03

Transfer learning makes the model train faster on the image set.

04

Class confidence makes model mistakes easier to inspect.

Specs

4,000 Pap smear images

Five-class medical image workflow

Reproducible preprocessing pipeline

Transfer-learning evaluation

Stack

PythonTensorFlowTransfer learningMedical imaging

View GitHub

Route monitor

scan window 240ms

WETH

USDC

ARB

WBTC

DAI

Pairs checked13,240+

Spread filterpass

Slippage guardon

RouteDEX A to B to C

Web3 automation

Flashloan Bot

Scans DeFi routes and filters out trades that are too risky or too slow to execute.

Features

01

Checks 13,240+ token pairs for route opportunities.

02

Separates scanning, validation, and execution into clear stages.

03

Filters by spread, gas cost, and slippage before acting.

04

Designed for automated checks instead of manual market watching.

Specs

13,240+ token pairs evaluated

Execution guards for timing and slippage

Separated discovery and execution paths

Stack

Solidity / EVMDeFiRoute scanningAutomation

View gist

/stack

Technical toolkit

$cat stack.json

core/TypeScript · Python · SQL

product/React · Next.js · Tailwind

services/Node.js · FastAPI · PostgreSQL · Supabase

ai-voice/Gemini · Twilio

ship/Docker · Vercel · GitHub

/experience

Work experience

Rezolve.ai

AI Voice Engineer · Contract·2025

Helped build call-flow logic for intent recognition and call handling

Worked on the ITSM integration path for ticket creation and escalation

Contributed before the product moved into larger enterprise deployments

Voice AINLPITSM IntegrationConversational AILLM Orchestration

Contact

Let’s talk

If the work here lines up with what your team is building, send a note. Email is best. LinkedIn, GitHub, and my resume are right here.

Emailshreyansh.sharma01@student.csulb.eduOpen LinkedInlinkedin.com/in/sharmasshreyOpen ResumeOpen PDFOpen GitHubgithub.com/shrey1110-dotcomOpen