Umang Gupta — Build Log

Log 01 — Early Years

Root-cause thinking

My journey didn't begin with venture capital or startups. It began with understanding how complex systems behave — and how seemingly independent parts can work together to create extraordinary outcomes.

IIT BOMBAY
PROCESS OPTIMIZATION &
CONTROLS ENGINEERING

RELIANCE INDUSTRIES
EA TO PRESIDENT

ORACLE
ERP IMPLEMENTATION

IIM INDORE
MBA, GENERAL MGMT

A — Learning to Think in Systems

At IIT Bombay, I studied Process Optimization & Controls Engineering, where I learned to view the world as interconnected systems rather than isolated events.

Engineering taught me that behind every complex outcome lies a model waiting to be understood. This mindset of breaking problems into first principles has shaped every investment, company, and strategic decision I've made since.

B — Distributed Systems & My First Bitcoin

While at IIT Bombay, I became fascinated by peer-to-peer (P2P) networks and distributed computing long before they became mainstream — no central authority, every node equal, trust established between peers rather than handed down by a hierarchy.

Assembled multi-core CPU PC→MINED BTC→Added NVIDIA GeForce Card

In early days, I mined a few BTCs using my assembled PC while experimenting with the protocol — but later sold them when Bitcoin first crossed into $ double digits. Financially, it wasn't my best decision, but it taught me one of my earliest investing lessons.

Far more important than the price was the protocol itself. The idea that a decentralized network could establish trust without a central authority fundamentally changed how I thought about incentives, coordination, and scale.

Never sell your highest-conviction bets too early.

Today, the same principles power AI infrastructure, distributed computing, and modern venture ecosystems.

C — Probability Over Certainty

My Master's thesis focused on stochastic processes — the mathematics of uncertainty. Given enough historical data, very little is actually random; cause and effect can be recovered, and the probability of an outcome estimated.

Probability

Given enough data, uncertainty becomes probability.

One foundational idea, grown into many disciplines.

Still how I approach investing today: identify signal, quantify uncertainty, and make asymmetric bets.

D — Engineering Meets Business

Understanding systems wasn't enough. I wanted to understand how organisations scale.

OS Layer 1 — Engineering

Reliance Industries

Executive Assistant to the President (Commercial & Corporate Affairs) — redesigning port jetty logistics and turnaround operations at one of India's largest industrial sites.

Oracle

ERP implementation delivery, cross-functional execution — enterprise transformation.

IIM Indore

MBA, General Management. Unanimously elected Class Representative — highest pre-placement offer in the class.

OS Layer 2 — Business

This wasn't a career change. It was the next layer of the same operating system — learning how to build, lead, and scale organisations as effectively as the systems behind them.

Log 02 — 2015–2016 · Dubai

First build, first exit

First proof point: build a functioning commercial operation from zero, to a specification someone else would acquire.

LCAM
INDUSTRIAL ASSET
MANAGEMENT

FOUNDING MEMBER &
GENERAL MANAGER

Founding member, LCAM — industrial asset management and services, Dubai. Built commercial operations, finance, supply chain, and project delivery across multiple verticals, from zero.

Scaled to $11.6M revenue, $1.2M EBITDA. Acquired into a KKR-backed platform — Joulon. That acquisition became the entry point into the next chapter.

Log 03 — 2016–2021 · UK · USA · UAE · Qatar · Singapore · India · China · Kuwait

Investor-operator

Not only evaluation. Structured the deals, then operated the businesses those deals created.

As VP, Investment, Finance, Operations & Corporate Development at Joulon, a KKR-backed industrial platform, I structured the deals, then ran what they created.

AStructuredDeal flow

60+

opportunities analysed

15

closed · 11 majority + 4 growth

$254M+

capital deployed

LBOCarve-outSecondary recapAcquisitionMajority stake

$100M+ proprietary deal flow originated

BOperatedPost-acquisition

8employees at start

2,500+employees at scale

5

boards

14

companies

$200M+

platform ARR

22%EBITDA expansion

45%revenue CAGR

$212M

capex programmes managed

$8M+

procurement savings
(legal costs cut $2.3M)

650+

vendors under one supplier programme

120

person back-office team led

Capital markets fluency, proven at global scale. Then the operating parameters changed.

Log 05 — 2021–2026 · UK · USA · Germany · India

Venture builder

Same rigor, applied at build stage. The evaluator became the operator being evaluated.

Co-founder / VP, AI Product & Engineering across two UK deeptech/HealthTech ventures in parallel, raised early-stage & growth capital to build AI-driven medical imaging technologies from R&D through fundraising and commercialisation, across product, engineering, IP licensing and investor relations.

Deeptech · AI Diagnostics · Oxford Spinout

Sonosine — Patented Electromagnetic Acoustics (EMA) Technology

1 / 6Sound tags a spot

Sound tags a spot
Radio lights the area
The tagged spot answers differently
Filter isolates that spot
Repeat to paint an image
Perfectly aligned with ultrasound

Detects cancer, calcification and fluid that ultrasound alone misses.

Owned the AI product roadmap to combine ultrasound and radio wave signal data into MRI/CT-equivalent diagnostic-grade images — resulting in a ~4–8dB signal-to-noise improvement over single-modality baseline and enabling detection of cancer, calcification and fluid formation undetectable by conventional ultrasound.

Trained and validated deep learning models across local GPU workstations (prototyping) and AWS cloud infrastructure (distributed multi-GPU training) for both ventures.

How it works

Model Stack

Layer 1 — Image Fusion (CNN)

Fusing ultrasound and electromagnetic signal data into diagnostic-grade images — ~4–8dB SNR improvement over single-modality baseline.

Layer 2 — GAN Translation

Conditional GAN with PatchGAN discriminator converting fused data into MRI/CT-equivalent diagnostic images — ~0.70–0.80 SSIM against paired MRI ground truth.

Layer 3 — Classification

Transfer-learned ResNet/EfficientNet backbone distinguishing malignant tissue, calcification and fluid formations — ~82–86% sensitivity, AUC ~0.85 in early clinical validation.

Diagnostic-Grade Output

Deeptech · Whole-body Imaging

Spectroma — Ultra-Wideband Microwave Imaging

1 / 6A short pulse goes in

A short pulse goes in
It bends around what's inside
Sensors listen all around
Repeat from every angle
Physics-guided AI rebuilds it
Image in under 5 minutes

Sees through tissue in under 5 minutes, with no radiation.

Built 3 prototypes, raised non-dilutive funding and successfully licensed the technology to Neko, the preventive-healthcare company backed by Spotify founder Daniel Ek.

Owned the technical strategy for solving the inverse scattering problem in UWB microwave tissue imaging, prioritising reconstruction speed as a core product requirement and championing a physics-informed neural network approach to cut required training data.

How it works

Model Stack

Layer 1 — Inverse Scattering Reconstruction

Physics-informed deep learning delivering single-pass reconstruction in under 5 minutes — ~0.65–0.75 SSIM against phantom ground truth.

Layer 2 — Physics-Informed Training

Embedding electromagnetic wave-propagation constraints into model training — reducing required training data by an estimated 30–40% versus purely data-driven baselines.

Layer 3 — Classification & Signal Processing

Distinguishing malignant/benign tissue from dielectric-contrast signatures — ~80% sensitivity, AUC ~0.82, signal-to-clutter improved ~8–12dB.

2D → 3D Volumetric Roadmap

Healthcare · India · Bootstrapped

Pharmola

Co-founder, Chairman & Board Director (Part-time), 2022–Present. Built and scaled a pharmaceutical and medical-device partnerships platform with WHO-GMP/USFDA-certified manufacturers across India and the UK — from zero, entirely bootstrapped, personally owning partner economics, pricing tiers and channel strategy.

Founder-Operator→55+ Team · ~$350K ARR · 80% YoY→Chairman & Board Director

Scaled the portfolio to 30+ product partnerships across diabetes, pain management, gynaecology and nutraceuticals through strategic licensing, manufacturing and distribution agreements — owning partner segmentation, ICP definition, pricing architecture and exclusivity terms by category and geography.

Expanded into medical devices, sourcing and commercialising global partner technologies (Edge AI hearing aids, smart rings, smart cuffs, glucose monitoring, non-contact optical screening) for the Indian market. Grew the team to 55+, scaling revenue to ~$350K ARR at 80% YoY growth and 15% net margins.

Transitioned from founder-operator to Chairman & Board Director as the business scaled, with appointed CEOs now running day-to-day commercial execution while retaining strategic oversight of partnerships and capital allocation.

Log 06 — 2026–Present · UK · India

Board seats & capital

The same evaluator's eye, redirected: board seats, applied AI, and capital deployed as principal — not employee.

Edge AI · Hearing Tech

Sonotu

Providing board-level product, technical and commercial leadership for the company's Edge AI hearing-aid platform, spanning £2M seed+ financing preparation, scale-up, India commercialisation agreements, product roadmap and regulatory pathway.

£2M seed+ financing prep · India commercialisation agreements

Current Focus

Iron Ventures

Building a suite of AI-native products for the venture ecosystem, spanning founders, investors and evaluators, combining LLM-based product architecture with hands-on GTM ownership.

Deploying his own capital, alongside a network of family offices and strategic investors, into breakthrough science and deeptech.

01Fundraising automation

Project Auto

Designed and built an end-to-end fundraising automation system: a scraper layer that sources and enriches investor data, a knowledge graph of investor mandates and holdings, and a ranker that scores investor-deal fit.

LangGraph multi-agentOutreachInbox WatcherCadenceDraftingCompliancePipeline Update

Master agent orchestration · human-in-the-loop approval before anything is sent

02Pitch evaluation

PitchIQ

Built a proprietary reinforcement learning model delivering investor-grade pitch evaluation across 10 scored dimensions.

500+pitches evaluated

100+founders

2accelerator partners

40%D30 retention

Within 12 weeks of launch · vs. a 20–35% B2B SaaS benchmark

03Due diligence

DDIQ

Built a self-improving diligence model that ingests unstructured data-room files and folders and generates:

Categorised Investment Memoranda
Executive summary
Dashboards
Risk register
Red-team analysis

ironventures.ltd →

I build what
I believe in.

Root-cause thinking

First build, first exit

Investor-operator

"My mother was diagnosed with cancer. That moment redirected everything."

Venture builder

Sonosine — Patented Electromagnetic Acoustics (EMA) Technology

Spectroma — Ultra-Wideband Microwave Imaging

Pharmola

Board seats & capital

Sonotu

Iron Ventures

Project Auto

PitchIQ

DDIQ

Define the seat.
Let's check the fit.

Root-cause thinking

First build, first exit

Investor-operator

"My mother was diagnosed with cancer. That moment redirected everything."

Venture builder

Sonosine — Patented Electromagnetic Acoustics (EMA) Technology

Spectroma — Ultra-Wideband Microwave Imaging

Pharmola

Board seats & capital

Sonotu

Iron Ventures

Project Auto

PitchIQ

DDIQ

Define the seat.Let's check the fit.

Define the seat.
Let's check the fit.