Checkpoint 01 of 10. Senior full-stack engineer and product lead
Dhruv Gopani
I build AI products for regulated work: medical records, legal cases, payments, licences. Products that have to be right before they are fast.
Surat, India. Remote, full overlap with Europe, three to four hours with US East. Open to contract, employer of record or relocation. Currently leading MedChron at Omnis AI.
Product leadRoadmap, specifications and priorities for MedChron at Omnis AI, and the core features behind them.
AI you can citeEvery extracted fact carries the page it came from, or it is dropped. Evaluations before a prompt ships.
Ships end to endTypeScript from NestJS to Next.js, PostgreSQL and MongoDB, AWS and Kubernetes, tests included.
Checkpoint 02 of 10. Why hire me
What you get when you hire me.
Three and a half years across a solutions role, a product studio and a US legal-tech startup taught me one thing: features are cheap, trust is expensive. I build for the second.
Product thinking, not ticket taking
I own the roadmap and write the specification before the code. Every feature gets an acceptance cycle with the people who will use it, and a QA round is not done until the tracker says so.
AI features that survive scrutiny
Citations or nothing. Deterministic guards ahead of the model. A replay evaluation before any prompt change. Cost per run known before launch. This is how MedChron went from a demo to a product attorneys rely on.
Regulated-data habits by default
Tenant scoping, rate limits, audit trails and identifier scrubbing go in on day one, not after the first security ticket. I wrote the secure-coding rules my current team reviews every change against.
The whole feature, delivered
Backend, frontend, data, deployment and tests from one engineer who has shipped six B2B platforms and two products of my own. No hand-offs, no gaps between layers.
The first thirty days on your product
Read the code and the open tickets before I propose anything. Write down the three biggest risks I see and agree them with you.
Ship one visible fix in week one, so the team sees how I work before the larger changes land.
Put an evaluation or a test around the riskiest path, usually an AI output, a payment or a permission check, so we can change it without fear.
Set a weekly demo. Working software, not slides, and a written note of what changed and why.
Checkpoint 03 of 10. Experience, 2023 to now
Experience.
Omnis AISenior Full-Stack Engineer, Product Lead for MedChronJanuary 2026 to now
Remote. US legal-technology company building an AI practice platform for personal-injury law firms.
Lead MedChron end to end: roadmap, specifications and priorities agreed with leadership, core features built with a distributed engineering team.
Rebuilt the deduplication and extraction stages of the Gemini pipeline so a file from many providers comes back as one clean timeline, with deterministic merge guards ahead of the model.
Introduced page-level citations for every extracted fact, with a resolver that checks each reference against the record and drops anything it cannot verify.
Took Chi, the in-product assistant, from a tool-calling prototype to release candidate: cited patient-scoped answers, PDF export, three QA rounds.
Closed every security finding raised against document and entity routes, enforced tenant scoping and rate limits, and maintain the secure-coding SOP the team reviews against.
Built the append-only prompt registry with versioning and restore, added PHI scrubbing to the extraction prompts, and integrated MedChron with the CasePro legal CRM and the shared identity service.
The Coding StudioFull-Stack DeveloperJune 2024 to December 2025
Surat, India. Product studio shipping SaaS for small businesses.
Built Kwik Media, a photo-album SaaS for studios: React, Redux and TypeScript on a Node backend, MySQL on RDS, S3 and CloudFront, Kafka queues and push notifications. Live at kwikmedia.in.
Built Delta Wealth, an AI personal-finance app on Next.js, Prisma and Supabase, where Gemini reads receipts so budgets stay current without manual entry.
Ran Node microservices with Kafka and Redis on AWS (EC2, ECS, Lambda) with CI/CD pipelines.
OTPlessSolution EngineerFebruary 2023 to October 2023
Delhi. Passwordless authentication company.
Integrated WhatsApp-based passwordless login into client applications across React, Next.js and Node stacks, replacing SMS one-time passwords and their cost and delay.
Checkpoint 04 of 10. Omnis AI, January 2026 to now
MedChron: a case file in, a chronology the attorney can trust out.
MedChron is the medical-records product of a US legal-technology platform for personal-injury law firms. A case arrives as thousands of pages from many providers: emergency room notes, orthopedic visits, therapy logs, imaging, pharmacy records, bills.
MedChron reads all of it and produces a coded chronology, the dated story of what happened to the patient, with every fact tied to the page it came from. Attorneys use it to understand the injury, build the demand and answer questions about the history without reading the file again.
My role: product lead and senior engineer. I own the roadmap, write the specifications, set priorities with leadership, and build the core features with the team.
Illustrative screen drawn for this portfolio. Not a product screenshot. Sample data.
Chronology pipeline
Deduplicate files, OCR pages, extract facts with Gemini in stages on background workers, merge duplicates with deterministic guards, cite every fact to its page.
Chi, the assistant
Answers questions about one patient with cited facts from that file only, and exports the answer as a PDF the case team can share.
Bills and injuries
Reconciles medical bills against the record and validates injury codes against ICD-10 body parts, so the numbers and the diagnoses line up.
Integrations and portal
Syncs with the CasePro legal CRM with retry and an outbox, embeds by matter, exports combined PDFs, and gives clients a hardened portal to share records.
Why attorneys can trust it
Every fact cites its page or is dropped. Unverifiable output never reaches the attorney.
Prompts live in an append-only registry with versions and restore, and are replayed against known records before they ship.
Patient identifiers are scrubbed before the models see the text. Tenant scoping and rate limits hold on every route.
NestJSMongoDBGemini 2.5ClaudeBullMQRedisSocket.ioReactAWS S3Kubernetes on EKSBetter-AuthPrometheus
Checkpoint 05 of 10. Projects
Projects.
Two products of my own, two shipped at the studio, and the site you are reading.
Own product, 2026
AgentFlow
Teams of specialized agents (researcher, writer, reviewer, coder) run as a dependency graph and stream to the browser. Each step is saved before the next starts, so a failure halts the run without losing the work. Free, Pro and bring-your-own-key plans on Stripe, with keys encrypted at rest and limits enforced in the service layer.
A 59-node n8n workflow that researches a trend, writes the script, generates images and a voiceover, renders the video and uploads it to YouTube without anyone touching it. Three quality tiers priced at their real cost, and a fallback provider at every step so one outage does not stop a render.
n8nClaudeReplicateElevenLabsShotstackYouTube API
The Coding Studio, 2024 to 2025
Kwik Media
Photo studios had no clean way to deliver and sell albums. Kwik Media lets a studio upload a shoot, share it with the client, and sell prints and albums from the same page. Live in production.
ReactReduxTypeScriptNode.jsMySQL on RDSS3 and CloudFrontKafkaFCM
A personal-finance app where you photograph a receipt and Gemini reads and categorizes the transaction, so the budget stays current without manual entry.
Next.jsPrismaSupabaseGemini
Own work, 2026
This site
A React Three Fiber scene that renders only on demand, compiles every shader before it fades in, and holds 35 frames a second on a laptop with integrated graphics. Tested across Chromium, Firefox and WebKit, and on phones.
Next.js 16React Three FiberthreezustandPlaywrightVitest
Checkpoint 06 of 10. Delivered platforms, 2024 to 2026
Six platforms delivered, 2024 to 2026.
Multi-tenant B2B systems for real operations: wholesale trade, recycling, aviation licensing, payroll incentives, food supply, crypto exchange.
Wholesale quotation platform.Island wholesalers stopped traveling shop to shop: retailers request quotes, prices stay private per quote, payment and installment options are set per island, drivers confirm delivery by QR scan.NestJS, Prisma, PostgreSQL, Next.js, TanStack Query
Deposit-return recycling.Government sets products and prices, collectors pay the public for returns, finance settles verified claims, every step permissioned and audited.NestJS 11, Prisma 7, PostgreSQL, Next.js 16, Zod
License lifecycle.An aviation regulator off paper: apply, inspect, pay, issue, renew. Three license types, PDF certificates, cron-driven renewal notices, four permission roles.Express 5, Prisma 6, PostgreSQL, Next.js 15, node-cron, jsPDF, AWS S3
Incentive engine.Disputed spreadsheet payouts replaced by a configurable slab engine fed by attendance and customer feedback, 16 data models, soft deletes and a full audit trail across locations.Express 5, Prisma 6, PostgreSQL, Next.js 16, Joi, AWS S3
Food-surplus marketplace.Two-sided marketplace connecting surplus-food sellers with buyers. Dual payment rails, real-time order updates, map-based discovery, seller and admin dashboards.Next.js 14, Payload CMS, tRPC, MongoDB, Stripe, Razorpay, Mapbox, Socket.io
Exchange backend.Real-time order matching over WebSocket for BTC, ETH, TRON and XRP, KYC document workflows, funded deposits and cron-based settlement.Express, MySQL, Socket.io, web3.js, bitcoinjs-lib, tronweb
Checkpoint 07 of 10. Skills and how I work
Skills, all of them shipped to production.
Languages
TypeScriptJavaScriptSQL
Frontend
ReactNext.jsReduxTanStack QueryTailwind CSSMaterial UIshadcn/uiFramer MotionReact Three Fiber
Core team, Google Developer Student Club (2020 to 2021).
Education
Bachelor of Engineering, Computer Engineering. Sarvajanik College of Engineering and Technology (Gujarat Technological University), 2019 to 2023. CGPA 8.74/10.
Checkpoint 09 of 10. Writing
Writing.
Every fact cites its page, or it is dropped.How MedChron turns thousands of pages of medical records into a chronology an attorney can use, and how we measure whether it is right.7 minute read. Draft.
Email is fastest. I answer within a day, and I am happy to walk through any of this on a call.
Surat, India. Remote, full overlap with Europe, three to four hours with US East. Open to contract, employer of record or relocation. Currently leading MedChron at Omnis AI.
