Projects

I build them myself.

Most product people write the spec and hand it over. I keep opening the editor anyway. Ten things below: a couple for work, one for the MBA, the rest because the problem was annoying me. Every one of them is real software you can open right now.

github.com/Maddddyy ↗ Ten builds 2023 → today

effectus.madhvendra.com

01 · Effectus Research

Strategic Validity System

Built solo, methodology with founder Brian Mooney MBA internship · 2026

Software engineering keeps Architectural Decision Records, so anyone can go back and see why a choice was made. Strategy keeps nothing. Boards commit, the assumptions underneath quietly stop being true, and nobody notices until it is expensive. This is the decision record strategy never had.

CARMR framework Drift detection Immutable versions

0 to 100

Integrity score

Weighted parts

80+

Strong governance

Open the live system ↗

livecoaching.app

02 · macOS app

LiveCoaching

Solo build · app, site, and launch film 2026

Interview advice is mostly generic and never measured. This watches a real rehearsal and tells you what actually happened: how fast you talked, where your eyes went, what you said in the first minute. Every note points at a line you really said. Nothing leaves the Mac.

Runs fully offline Camera optional Evidence-linked

100%

Local, internet off

88.5s

Launch film

Signed

Invite-gated build

See the product ↗

Virtual Maddy

LISTENING04:12

VOICE

Pace 148 WPM ✓

Vol 62%

Pause 18%

Fillers: uh×2 like×1

COACH

8.2

CLARITY

7.5

CONFIDENCE

7.8

PERSUASIVE

Land the number first, then explain it.

Confident Eye contact 8.4 Slide read

03 · Personal AI

Virtual Maddy

Solo build · macOS menu bar app 2026

A rehearsal coach that sits in the menu bar and listens while I present. It holds me to a 120 to 160 word pace, counts the ums and likes, and docks me for hedging with “I think” and “sort of”. It watches my face for eye contact, reads the slide on screen, and every few seconds returns the one thing to fix next. I built it because my own presenting needed the work.

Pace, fillers, hedging Eye contact Reads your slides

120 to 160

Target word pace

Live channels: mic, cam, screen

Every 4s

Camera read

See its sibling, the career agent ↗

aceitchamp.com

AceItChamp features: AI tutor, smart learning path, assessment suite, multilingual support, offline learning

04 · Founder, non-profit

AceItChamp, with ShikshaSathi inside it

Founder · concept, product, and pilot 2023 to 2025

AI tutoring launches in English first, which quietly writes off the students who do not study in it. ShikshaSathi began as a Hindi-first tutor fine-tuned on the national curriculum, with every answer checked before a student could rely on it. Ola backed it. AceItChamp grew out of that work, and ShikshaSathi is still the tutor inside it.

Hindi-first Llama 3, fine-tuned Backed by Ola Offline mode

Institutes piloted

Weekly

Classroom feedback

Open the platform ↗

afterward.vercel.app

05 · Health AI · Award winner

AfterWard

With four teammates · Startup Fast Track Edinburgh · Nov 2025

After surgery people go home and drop out of sight until something goes wrong. AfterWard keeps checking. It calls and messages patients through recovery, grounds its triage in clinical guidance, and hands over to a clinician the moment it should. Built around three NHS surgical pathways in Edinburgh.

Voice and text agent Clinician escalation Audit trail

Winner

Most Viable Business

NHS pathways

Open the prototype ↗

companyf.vercel.app

06 · Built for my MBA cohort

Company F Dashboard

Solo build, for the team on the table Edinburgh · 2025

Our operations module ran a live manufacturing simulation on a two hour fifteen trading window. Spreadsheets could not keep up with it, so I built the team a control room: timers pacing the rush orders and the cut-off, marketplace specs mirrored from the brief, and the fold and stencil rules on standby for whoever needed them.

Live timers Order board Finance and inventory

2h 15m

Trading window

Live

Used during play

Open the dashboard ↗

localhost:3000

Role Experience Location Education +5 more

Estimated salary

95% confidence range

Feature contribution, SHAP

Experience

Role

Location

Education

07 · Full stack

AI Salary Predictor

Solo build · model, API, and interface 2026

Salary tools hand you a number and no reasoning. This one shows its work: which inputs pushed the estimate up, which pulled it down, and a confidence range instead of one false-precise figure. The same discipline I want from any AI system with consequences attached.

Random Forest SHAP explainability FastAPI + Next.js

96.56%

R² accuracy

200K+

Training records

Input factors

Read the code ↗

Also built