// profile matched

Vinayak Sharma

Senior Full-Stack Product Engineer — Flutter · Computer Vision · GraphQL

6+ years at the awkward edge of hardware, domain and software. Computer vision and embedded Linux on one side; multi-tenant GraphQL and offline-first field clients on the other. Sole architect of a four-product instrument-software suite running hardness testers and metallurgical microscopes in production labs.

Portrait of Vinayak Sharma framed by a red targeting reticle with the caption “Profile matched”.

// 01 About

I build whole products, not slices of them — the measurement maths, the interface an operator learns once, the service it syncs to, and the Linux box it boots on. The work that suits me best is the kind where a wrong number costs money, a certificate or a person, so the tolerances, the calibration and the audit trail are part of the build rather than a phase somebody funds later.

Since 2024 I've been Lead Developer at Oz SoftTech, sole architect and engineer of the Multitek suite: four Flutter products across desktop and embedded Linux, driving Vickers/Brinell hardness testers and metallurgical microscopes in production labs. Before that I led engineering for Intrisafe GmbH in Hamburg — an eight-product workplace-safety platform covering a worker's whole path onto a German industrial site — with a team of 3 engineers. B.Tech in Electronics & Instrumentation from VIT Vellore (2019–2023).

Programming

  • Dart
  • TypeScript
  • Python
  • Go
  • C#
  • SQL

Client

  • Flutter
  • macOS · Windows · Linux
  • iOS · Android · Web
  • BLoC
  • Custom-painter canvases
  • Offline-first sync

Vision & data

  • OpenCV
  • Pure-Dart image pipelines
  • ASTM / ISO metrology
  • ML Kit
  • TensorFlow

Platform

  • Node.js
  • GraphQL
  • Prisma
  • MongoDB
  • gRPC & protobuf
  • Appwrite
  • Firebase
  • Multi-tenancy
  • Signed documents

Systems

  • Embedded Linux
  • Wayland
  • Raspberry Pi · GPIO
  • Docker
  • Kubernetes
  • CI/CD
  • OTA & licensing

// 02 Experience

  1. Lead Developer · Oz SoftTech Private Limited

    Jan 2024 — present · Remote
    • Sole architect and engineer of the Multitek suite — four Flutter products across desktop and embedded Linux, operating hardness testers and metallurgical microscopes behind one calibration model, report format and operator interface.
    • Built the shared platform: an in-house camera plugin putting full-resolution industrial USB/GigE feeds into zero-copy GPU textures, plus gRPC licensing with per-app identity, TOTP exchange, clock-tamper defence and OTA updates.
    • Ported the suite onto embedded Linux — a Raspberry Pi touch panel booting straight into a Wayland kiosk, GPIO over dart_periphery, and a supervised firmware bridge that self-restarts so a fault never strands the operator.
    • Built Intrisafe's gatehouse induction kiosk — an unattended Surface-tablet terminal in 26 languages, no login, issuing signed certificates verified against a public key; OTA deployment covers 25+ devices.
    • Lead Limse, a hyperlocal social-commerce app on Android, iOS and web from one Flutter codebase — offline-first caching, in-app video editing and Google ML Kit for on-device content understanding.
  2. Lead Software Engineer · Intrisafe GmbH

    Jun 2021 — Dec 2023 · Remote (Hamburg, DE)
    • Led engineering for a Hamburg workplace-safety SaaS whose platform spans eight products — covering a worker's whole path onto a German industrial site, from gatehouse induction to the safety-office console — leading a team of 3 engineers.
    • Designed the single data model behind all of it — employees, sites, work equipment, qualifications, instructions, inspections, incidents — over GraphQL / Prisma / MongoDB with subscriptions, so a worker is one identity whether the fact was entered at a terminal, on a phone at a scaffold, or in the office.
    • Built a multi-tenant control plane over company, user and role records: one deployment serving every customer their own branded instance, on Docker and Kubernetes with CI/CD collapsing a manual release into one automated deploy.
  3. Freelance Full-Stack Engineer · 13 projects

    Apr 2020 — Jan 2023 · Remote
    • Delivered 13 full-stack projects across industries and regions — Flutter, Firebase, Node.js, GraphQL, .NET, Unity, TensorFlow, Python — shipping to embedded, desktop, mobile and web, each owned from scoping to handover.
  4. B.Tech, Electronics & Instrumentation · VIT Vellore

    2019 — 2023

// 03 Projects

Client work is private (no public repos) — each case study below covers the problem, my solution and the stack. Contact me for a guided test run of any app.

Multitek — the metrology bench, in software

Oz SoftTech · sole architectFlutter desktop · OpenCV · gRPC · embedded Linux

A metallurgical lab does not own one machine. It owns a hardness tester, a compound microscope and a stereo zoom scope — and until now each came with its own software, its own units and its own export format. These four products share one calibration model, one report format and an interface an operator only has to learn once. Nothing is ever reported in pixels, every result carries the overlay that produced it, and the certificate comes out of the lab's own template. Offline, licensed per machine.

  • 4products
  • 20measurement modes
  • 10ASTM / ISO methods

Hardness Works

Desktop · hardness testing

It replaces the eyepiece and the spreadsheet. The app owns turret, load and dwell on a Vickers/Brinell tester, watches the microscope camera and detects the indentation diagonals itself (OpenCV) — with a manual override for low-contrast etches, because an ambiguous edge should never be a black box.

  • Derives HV/HBW off the measured impression, with ASTM E140 conversion across to HRC
  • Load calibration is a per-step table stored with the machine profile — a real load cell drifts off its nominal curve
  • Trend, distribution and case-depth charts roll up across a session into a templated certificate
Hardness Works measuring an indentation with detected diagonals overlaid on the live camera image

Image Works

Desktop · metallurgical analysis

A metallurgist has to answer in numbers — how coarse the grain, how clean the steel, how deep the decarburized layer. Traditionally that means holding the eyepiece view against a printed comparison chart and writing down the closest match. Image Works measures it instead.

  • 10 ASTM/ISO methods implemented down to the annex rules — E112 grain size, ISO 4967 inclusion rating with stringer merging, E1245 phase fraction, A247 nodularity, E1077 decarburization
  • Pipelines are pure Dart with no widget imports, so the science is unit-testable on its own
  • Every result ships with the overlay that produced it — the operator can see exactly what was counted, and disagree
Image Works grain-size analysis with the measured overlay drawn over a micrograph

Image Lite

Desktop · calibrated measurement

The product most benches actually need: they already know what they are looking at, what they need is a defensible dimension and a document. Same codebase as Image Works, shipped as the measurement-only SKU.

  • 20 measurement modes on one canvas — distances, parallel spacing, angles, circles, arcs, freehand, annotation — each labelled on the image and carried into the report
  • Per-objective µm/pixel calibration profiles, so nothing is ever reported in pixels
  • Panorama stitching and focus stacking run as queued external processes, so a long stitch never freezes acquisition
Image Lite measurement canvas with labelled distances and angles over a live stereo-zoom feed

Hardness Works Core

Embedded Linux · standalone tester

The same tester on embedded hardware, with no PC beside it: a Raspberry Pi touch panel booting straight into a Wayland kiosk.

  • GPIO over dart_periphery for turret, load and dwell control
  • A supervised firmware bridge that self-restarts, so a fault never strands the operator mid-test
  • Shares the calibration model, report format and OTA channel with the desktop products

Intrisafe — a workplace-safety platform, end to end

Intrisafe GmbH · engineering leadFlutter · Node.js · GraphQL · Prisma · MongoDB

Eight products covering the whole path a worker takes onto a German industrial site: the terminal in the gatehouse, the training they sit through, the phone the site team carries, the certificate anyone can verify, and the console the safety office runs it all from. Safety only counts if it can be proven later — so every product writes to the same record. An induction taken at a gatehouse terminal is the same fact the safety office reads in the console, the supervisor sees on a phone at the scaffold, and a client verifies months afterwards.

  • 8products
  • 26languages
  • 3engineers led

Intrisafe App

Web · the safety-office console

Employees, instructions, qualifications, medical records, projects, tasks, inventory, work equipment, inspection reports, incident analysis, e-learning and a document vault — one system where the job is usually a folder of spreadsheets.

  • The spine is the qualification lifecycle — active, expiring, expired — and that state generates the task queue, rather than somebody remembering
  • Templated inspections capture defects, photo evidence and an on-device signature, then export to PDF
  • GraphQL subscriptions over WebSockets keep every open dashboard live
Intrisafe safety-office console showing the qualification register

Intrisafe Backend

Service · the shared record

One schema behind all of it — employees, departments, work equipment, qualifications, instructions, inspection reports, incidents, courses, vault items and places — so a worker is one identity across every client in the platform.

  • GraphQL over Prisma and MongoDB, multi-tenant with role- and permission-based access control
  • Subscriptions for the events that have to arrive rather than be polled for: an inspection filed, an incident reported, a task assigned
  • Docker and Kubernetes with CI/CD pipelines for highly available deployment
Intrisafe backend schema and service overview

Intrisafe Companion

iOS · Android · the field client

The phone client for people on a site rather than at a desk: inspections with their templates, reports and defect closeout, the work-equipment and maintenance register, the document vault, incidents, projects and employee records.

  • Drafts are held on device and sync when signal returns — a scaffold inspection in a basement is exactly where the connection is not
  • On-device SQLite store synced per-tenant, per-user; uploaded drafts are validated for integrity by the backend before merging
  • QR-scan-based navigation straight to an asset or an inspection
Intrisafe Companion inspection screens on a phone

Intrisafe Learning

Web · desktop · training

Not a catalogue somebody browses — an assignment raised by a gap in the qualification register. A course mixes video, narration, PDF handouts and images in one player, then a timed test with a pass mark.

  • Passing issues a certificate that writes back to the qualification that generated it, so the compliance gap closes without anyone re-keying anything
  • Renders five content types — image, audio, video, PDF, webview — in one course flow
  • Token sign-in links, because asking a site crew to remember a password is how training does not get done
Intrisafe Learning course player with mixed media content

Intrisafe Kiosk

Terminal · gatehouse induction

A subcontractor walks up to an unattended Surface tablet, picks one of 26 languages, types their name and company, watches the site induction and takes the test. No account, no login — the terminal opens an anonymous session that expires in an hour.

  • Results are signed so the server can prove which terminal issued them
  • The induction video has a countdown and no scrub bar, because an induction is not something you skip
  • Branding, greeting, languages and video all come from the module record — one binary serves every customer, and the URL decides which
  • OTA deployment covers 25+ devices in the field
Gatehouse induction kiosk language-selection screen

Intrisafe Kiosk Backend

Service · terminals, video, certificates

One Node deployment serves every customer their own terminal URL, streams the induction video, and admits certificates only if they verify.

  • Video goes out over range requests in 1 MB chunks from a local disk cache, invalidated by size against the origin — terminals sit on site links, and pulling a video per visitor is not viable
  • Certificates arrive signed and are verified against a public key before anything is written, so a forged or replayed result fails there rather than in the database
Kiosk backend service overview

Intrisafe Kiosk Analysis

Web · training analytics

One login per training module, then what the safety team wants to know about it: who finished, what they scored, which languages they chose, and where people fall out.

  • Reads the certificate records the terminals write rather than a separate event stream — the record that proves the induction happened is the record the numbers come from, so there is nothing to reconcile
Kiosk analytics view showing completions, scores and drop-off

Intrisafe Admin

Web · multi-tenant control plane

Deliberately narrow: the companies on the platform, the users inside them, and the roles those users hold. One deployment serving every customer their own branded instance.

  • A tenant record carries what support actually needs — trade association and registration number, employee and record counts, join date — so an operator can settle an account query without opening the customer's own instance
Intrisafe admin console listing customer companies and their users

Also built Intrisafe Verify — every report the platform generates carries a unique checksum, and this public-facing app confirms a forwarded PDF has not been altered.

Limse

Oz SoftTechFlutter · Android · iOS · Web · ML Kit

Problem Hyperlocal sellers live on social media, but the discovery and the buying happen in two different apps — so the purchase leaks away between them.

Solution A hyperlocal social-commerce app that puts a full social feed and the shop in one place, on Android, iOS and web from a single Flutter codebase.

  • Offline-first caching so the feed and catalogue survive a dead connection
  • In-app video editing for sellers producing their own content
  • Google ML Kit for on-device content understanding

SIGEHUB Web App

GEFAS mbHFlutter · Node.js · GraphQL · Object storage

Problem On-site safety audits were paper-driven; stakeholders had no central place to review results.

Solution A SaaS hub that records key safety metrics in digital forms during audits and serves condensed event summaries to stakeholders.

  • Performant object-storage pipeline for large volumes of photo and video evidence
  • Digital audit forms with map-based site views and protocol logs
SIGEHUB map view of audit sites

// 04 Contact

// primary connection routes