SYS.PROFILE — AI Systems Engineer

Ramanuj Ladda.

I architect multi-agent AI systems that leave the notebook and survive production — analytics, government workflows, mental health, commerce.

Computer Science @ COEP Technological University, Pune. Six products live. Every one of them is doing a job somebody used to lose weeks on.

View the work ↓ laddaramanuj@gmail.com
backlog.sim — interactive READY

Think you can beat my agents?

20 seconds. you vs. an incoming backlog. then my agents show you how it's done.

0

products in production

0+

live users served

0%

workload automated away

0

organisations shipped for

01 — Selected Work

Systems in production

Not demos. Not prototypes. Four platforms with real users, real uptime and numbers that moved — each one designed, engineered and deployed end-to-end.

FIG. 01 ANALYTAI.SYS SEMANTIC CATALOG — 1,247 TABLES MAPPED
/01 LIVE — INDUSTRIAL SAAS

AnalytAI

AI Analyst for Live Business Data

Every business question, answered in seconds.

A production SaaS where teams ask plain-English questions about live data and get streamed answers as charts, tables and metrics. A multi-agent system routes every query through specialised sub-agents, backed by a 3-tier agentic memory and a semantic catalog engine that maps 1,000+ tables with zero manual onboarding — replacing BI tools, analyst queues and spreadsheet exports.

0 +

organisations in production

0 s

time-to-answer, down from 6–8 hrs

0 +

tables mapped, zero onboarding

0 +

live data connectors

LangGraphLangChainQdrantFastAPIPostgreSQLReactTypeScriptDocker
Visit analytai.in ↗
FIG. 02 COEP CAS.SYS REVIEW CHAIN — 5 ROLES, 1 ENGINEER

APPROVED

3 WEEKS → <2 HRS

/02 LIVE — GOVERNMENT

COEP CAS

Government Faculty-Promotion Platform

A 100-year-old paper process, digitised solo.

Sole engineer on a government-commissioned platform replacing 12+ years of manual faculty promotion workflows at COEP Technological University. A 5-role review chain — Faculty → HOD → Dean → Registrar → Superadmin — cut submission time from 3–4 weeks of physical paperwork to under 2 hours, with final statistics formally submitted to the Maharashtra State Government.

0 hrs

submission time, down from 3–4 weeks

0 +

live applications from 200+ users

0 %

administrative time reduced

0 GB+

documents, zero data loss

HonoBunReactTypeScriptPostgreSQLDrizzleTanStackDockerCaddy
Visit cas.coeptech.ac.in ↗
FIG. 03 MANASMITRA.SYS VOICE SESSION — LATENCY 180MS
/03 LIVE — STATE INITIATIVE

Manasmitra

AI Mental-Wellness Platform · AMRUT Maharashtra

A companion that listens, 24 hours a day.

A state-initiative mental wellness platform serving 400+ production users across Maharashtra — crisis-aware chat with a parallel safety-triage agent, Gemini Live native audio-to-audio voice sessions, 50-question self-reflection paths, clinical assessments with severity banding, and a full admin intelligence layer with therapist verification.

0 +

production users across Maharashtra

0 /7

crisis-aware AI companion

0 -Q

guided self-reflection pathways

0

agents per conversation — care + safety

Gemini LiveLangGraphLangChainFastAPIPostgreSQLReactCloudflare R2Docker
Visit amrutmanasmitra.org ↗
FIG. 04 TRYON.SYS BATCH RENDER — 6/6 COMPLETE
◂ ▸
/04 LIVE — E-COMMERCE

TryOn

AI Virtual Try-On · A Byvent Product

See it on yourself, before it exists.

An AI-powered virtual try-on SaaS letting shoppers visualise clothes, sarees, jewellery and accessories on their own photos. A batch-generation engine renders 6 garment styles concurrently, with a drag-to-reveal comparison slider — live with e-commerce clients and lifting conversion by 25% on average.

0 %

average lift in conversion

0 ×

concurrent garment renders

0

e-commerce clients in production

0

categories — clothes, sarees, jewellery, more

React 19HonoBunCloudflare WorkersD1R2Better AuthDrizzle
Visit tryon.byvent.com ↗

02 — Experience

Where I've shipped

/01

Amrut Maharashtra

Full Stack Engineer

Shipped Manasmitra, a 24/7 AI mental-wellness platform serving 400+ users — Gemini Live voice sessions, crisis-aware triage, R2 audio streaming and an admin layer with automated therapist verification.

Aug 2025 — Present
/02

Neural Stream Automations

AI Automation Intern

Automated document verification across 33+ government schemes — an agent pipeline that downloads applicant PDFs, extracts and validates data against official records, cutting processing time by 95%.

Jun 2025 — Jul 2025
/03

GurujiWorld

AI Engineer Intern

Built AI automation reducing teacher workload by up to 95% — evaluation, lesson plans, content and OCR pipelines, plus microservices powering flashcards, mind maps and writing assistants.

Mar 2025 — Jun 2025
/04

RSquare Datavision

Software Intern

Built a cross-platform desktop tool converting natural language to SQL for MySQL/PostgreSQL — instant tables and insights for non-technical users, with a modular connector backend.

Oct 2024 — Mar 2025

THE WIDER ORBIT

Ventures & experiments

BRAND / STUDIO ↗

Byvent

The parent brand. A product studio shipping AI-native software — TryOn is the first product under the umbrella.

byvent.com
PLATFORM ↗

Digital Vaikuntha

A digital platform bringing culture and community online.

digitalvaikuntha.com
PRODUCT ↗

Picco

A consumer product exploring commerce experiences.

picco.co.in

03 — About

The operator

I'm a final-year Computer Science engineer at COEP, Pune — but the degree is the footnote. The record is six production systems across analytics, government, mental health and commerce, built end-to-end, mostly solo, all live.

My obsession is the gap between AI demos and AI infrastructure — systems with memory architectures, safety layers, semantic catalogs and multi-role workflows that hold up when real organisations depend on them. If a process wastes weeks of human time, I take it personally.

EDUCATION.RECORD VERIFIED

COEP Tech

BTech CS · 2023–2027

8.50 CGPA

Computer Science

99.80%ile

MHT-CET · State Rank Band

OPERATING PRINCIPLES

P.1

Production is the only benchmark

A system that works in a notebook is a claim. A system with 400 live users, daily backups and zero data loss is a fact. I only deal in facts.

P.2

Agents should replace queues, not people

The best automation removes the 3-week wait, the analyst backlog, the manual re-check — and hands humans the judgement calls that actually need them.

P.3

Own the whole stack

Schema to system prompt, Docker to DNS. When one person can see the entire machine, the machine gets simpler, faster and harder to break.

04 — Contact

Got a process worth automating?

I work with founders, institutions and teams who want AI systems that actually ship. If you're sitting on a workflow that eats weeks, or an idea that needs an engineer who finishes — write to me.

laddaramanuj@gmail.com Résumé / Dossier ↓ Replies within 24 hrs · Pune, India
01 GitHub laddaramanuj ↗ 02 LinkedIn ramanuj-ladda ↗ 03 X / Twitter @LaddaRamanuj ↗ 04 YouTube @laddaramanuj ↗ 05 Instagram laddaramanuj ↗