Working as a Junior Full Stack Developer at Mumbai Biocluster (ICT Mumbai Research Foundation), a Section-8 nonprofit based at ICT Mumbai in Matunga. The organization operates a single-use cGMP facility and works to translate early-stage startup and academic innovations into clinical-stage biotechnology, supporting startups, academic institutions, MSMEs, and industry partners across cancer, rare disease, and biologics research. Within this role I am the sole full-stack developer building and maintaining three production platforms for the organization: IndiaPharmaHub (IPH), a pharma/biopharma outsourcing marketplace; Yantra, a lab-instrument-booking and sample-testing marketplace built on the same architecture; and an internal HRMS covering attendance, payroll, KPI reviews, and recruitment for the organization itself.
02
Responsibilities
Platform ownership
Sole full-stack developer on three production Next.js platforms — authored 197 of the ~200 commits on IndiaPharmaHub, 125 of 125 commits on Yantra, and 77 of 77 commits on the internal HRMS — owning architecture, Prisma schema design, and feature delivery end-to-end from spec to deployment.
Designed and evolved three independent PostgreSQL schemas covering identity verification, catalog/taxonomy, quotations, messaging, reviews, audit logging, notifications, and webhook events (IPH/Yantra, 50+ models each), plus attendance, payroll, KPI reviews, and recruitment (HRMS, 41 models).
Backend and integrations
Integrated Razorpay for subscriptions/billing, AWS S3 for file uploads with presigned URLs and a media-proxy pipeline that pre-shrinks oversized images server-side, NextAuth v5 for authentication, and nodemailer for transactional email.
Built bulk data workflows on ExcelJS/xlsx for CSV/XLSX import and export (bulk product/account/instrument uploads with validation, duplicate prevention, and status/import history tracking), and PDF generation with pdf-lib.
Implemented GST/PAN-based verification workflows so sellers, buyers, labs, and clients are vetted before they can transact on either platform.
Quality and tooling
Wrote a custom dependency-graph-based "test-affected" Jest runner (madge-based, resolves the tsconfig @/* alias) so CI only re-runs tests transitively impacted by a change instead of the full suite — backs 520+ colocated test files on IndiaPharmaHub.
Enforced canonical Tailwind utility conventions (rounded-md over arbitrary values, has-[] selectors, standard easing) across both codebases for design-system consistency.
Made deliberate build-vs-revert calls under sole ownership — see the Yantra Custom Study note below — rather than shipping a duplicate feature path.
Infrastructure
Deploy and manage the platforms across AWS EC2 and Bluehost VPS instances, handling server setup, environment configuration, and release deployment myself alongside the application code.
A B2B marketplace connecting verified pharma, biopharma, and life-sciences sellers and buyers across India for outsourcing R&D, manufacturing, testing, packaging, regulatory support, and technology needs.
One account can hold a Seller profile, a Buyer profile, or both, with GST/PAN verification before a user can transact.
The platform received the India Innovation Catalyst Award from the DBT Secretary of India at #Biologics2025, before this rebuild. I rebuilt the application after joining in 2026 and am its sole developer (197 of 200 commits).
Legacy platform
Inherited a two-part codebase: an Express and Sequelize REST API with Socket.io chat (about 13,500 lines of JavaScript, 32 models, no automated tests) and a separate Next.js Pages Router client (about 44,000 lines) that called it over HTTP.
Audit findings: an environment file committed to the client repository, auto-increment IDs that made records guessable, no caching, rate limiting, or audit log, and email-only notifications.
The single EC2 server ran out of disk space and crashed; I restored it and assigned an Elastic IP.
New architecture
Replaced the two-service setup with one TypeScript application on Next.js 16 (App Router): 49 server-action modules, 120 library modules, and 12 API routes for uploads, webhooks, real-time events, health checks, and cron jobs.
Moved from Sequelize to Prisma 7 (adapter-pg) on PostgreSQL with 52 models, cuid identifiers in place of auto-increment IDs, and the pg_trgm extension for typo-tolerant search.
Replaced JWT middleware with NextAuth v5 sessions, with admin roles re-checked against the database, and split identity into User, Admin, SellerProfile, and BuyerProfile.
Added Redis for caching with TTL jitter, a Lua-script rate limiter, and pub/sub behind server-sent events, which replaced Socket.io for chat and notifications.
Moved file storage to AWS SDK v3 with presigned URLs, magic-byte upload validation, and a media proxy that resizes large images before next/image serves them.
Added idempotent Razorpay webhook handling, plan-based limits, and cron routes for plan reminders, onboarding reminders, and an email queue.
New features
Guest browsing of products, services, and providers without login, with in-place sign-up, SEO metadata, JSON-LD, sitemap, and llms.txt.
Role-specific dashboards with recharts trend charts, a seller Leads page, and hot or cold lead scoring based on recency, pipeline stage, buyer intent, and deal size.
Typo-tolerant search, location-based ranking (same city, then state, then country), and plan-based limits on directory results.
Wishlist, product and service reviews with rating aggregation, and quotation cards shown inside chat threads.
Pricing plans with a super-admin switch to enable or disable pricing enforcement per role.
Admin tools: audit log, support tickets, product approval workflow with seller review of admin-added products, bulk approve, reject, and delete, and admin-onboarded accounts.
Excel bulk upload for products, services, categories, product structures, locations, and accounts, with cascading dropdowns, per-row validation, and downloadable error reports.
Account deletion that archives records to separate tables first, and disposable-email blocking at sign-up.
Technical highlights
Wrote a re-runnable migration script that moved the legacy Sequelize database into the new schema and remapped S3 files to a new key scheme, with compare and backfill scripts to find duplicate or dangling records.
Fixed security issues found in code review, including stored XSS through SVG uploads, IDOR on support tickets, and non-atomic payment and approval updates.
Reduced layout shift on the products page (CLS 0.112 to 0.006) and removed about 150 KB of unused JavaScript through code splitting.
Built a Jest runner that runs only the tests affected by a change, using the import dependency graph; the repository has 520+ test files.
A sister marketplace to IndiaPharmaHub: connects Clients who need lab instrument bookings and sample/analytical testing with verified Labs, with Admin mediating every interaction — Client and Lab never see each other's identity directly.
Sole developer on this platform: 125 of 125 commits, full ownership of schema, RBAC, and every feature shipped.
Features
Granular, database-driven RBAC (Permission, PermissionGroup, Role models) replacing blanket admin/user role checks across the app.
Instrument catalog and booking with availability blocks, sample intake and tracking, a quotation-to-purchase-order-to-study-plan pipeline, invoicing, shipments, and lab reports with separate client-facing and internal-traceability renders from one underlying record.
Admin-in-the-middle messaging (CLIENT_ADMIN / LAB_ADMIN / SUPPORT thread types, deliberately no direct client-lab channel), escalations, reviews, wishlist, and notifications.
URL-synced list filters (search/status/date/category/sort/page), lab-catalog bulk-upload templates generated as real .xlsx files, and error boundaries with loading fallbacks to stop blank-screen navigations.
Technical highlights
Built a full AI-powered "Custom Study" intake subsystem (new thread type, intake model, assistant stub, admin queue) and then deliberately reverted it the same day after determining the existing quotation-to-study-plan pipeline already covered the same use case — verified zero database rows referenced the removed schema before dropping it, rather than leave two competing concepts in the app.
Iteratively simplified the Instrument model as the pricing-tier design matured — removed fixed pricing, structure, subcategory, and capacity/location fields — each change backed by its own Prisma migration.
Replaced full-page refreshes with targeted component updates and added granular RBAC enforcement in place of blanket admin checks.
An internal HR management system built for Mumbai Biocluster itself, not client-facing: attendance, leave, payroll, KPI reviews, recruitment, assets, announcements, training, and company policies in one platform.
Sole developer: 77 of 77 commits, 41 Prisma models spanning attendance/regularization, leave balances, payroll runs and salary slips, KPI cycles/assignments/reviews, job postings/applications/interviews, assets, internal chat, tasks, reimbursements, and job referrals.
Features
Attendance with session tracking and a regularization workflow, leave balances, payroll runs with generated salary slips, and an audit log across sensitive actions.
A full recruitment pipeline (job postings, applications, interview scheduling) and a KPI cycle system with templates, assignments, reviews, and comments.
Internal chat with group messaging and file sharing, task assignment with comments, company announcements with read-tracking, training enrollment, and a company-policy/holiday calendar with read-tracking.
Asset management with an asset log, reimbursement requests, and an employee job-referral flow.
Technical highlights
Drove a sustained frontend performance pass: code-split non-default tabs across KPI, Reports, Settings, and Policies; disabled Next.js Link prefetch on always-visible nav and dashboard links; inlined critical CSS to remove a render-blocking stylesheet request.
Seeded the root layout, dashboard cards, and multiple context providers server-side to kill client-side loading flashes, and fixed a session-loading race that was wiping SSR-hydrated context data.
Closed SSE connections on pagehide so pages qualify for the back/forward cache, and gave SSE reconnects their own rate-limit bucket separate from normal API traffic.
Used BullMQ and node-cron for background jobs (payroll processing, notifications) alongside Redis-backed caching.
06Event
PCG 2026 Symposium
The organising team on stage at PCG 202601 / 07
Overview
The Peptides & Complex Generics Symposium (PCG 2026), organized by Mumbai Biocluster and ICT Mumbai, ran on 28–29 September 2026 at Aurika by Lemon Tree Hotels, Mumbai: 400+ delegates and 70+ speakers working through peptides, oligonucleotides, GLP-1s, and the Billion Dose Challenge.
I was part of the Mumbai Biocluster team running the event, which put me in the same room as the people IndiaPharmaHub and Yantra are built for.
My role
Managed travel for speakers during the event.
Talked with speakers and industry leaders about IndiaPharmaHub and Yantra and asked for their views on both platforms, collecting product feedback first-hand instead of through a spec.
Built new connections across industry, academia, and the regulatory side of pharma.
Takeaways
Users are not a spec document: a few minutes of honest feedback from the people a platform serves beats weeks of guessing.
In pharma, trust is the product. Nobody hands their research to a platform they don't trust, which is why verification sits at the core of IndiaPharmaHub and Yantra.
Scaling medicine to a billion doses is not only a science problem; it also needs systems that connect labs, CDMOs, startups, and industry.
07
Conclusion
Since joining Mumbai Biocluster, I've been the sole full-stack developer behind three production platforms built for ICT Mumbai Research Foundation — IndiaPharmaHub, a pharma outsourcing marketplace; Yantra, a lab-instrument-booking and sample-testing marketplace; and an internal HRMS running the organization's own attendance, payroll, and recruitment operations. Owning all three end-to-end has meant designing independent Postgres/Prisma schemas, wiring payments, storage, and verification workflows, building my own CI test tooling, driving a real frontend performance pass, and making the judgment calls that come with sole ownership — including building then deliberately reverting a redundant AI subsystem once a simpler existing flow was shown to already solve the problem. Outside the codebase, I was part of the team behind PCG 2026, where I heard what speakers and industry leaders need from IndiaPharmaHub and Yantra directly from them. The role has deepened my experience designing multi-sided, trust-and-verification-driven marketplaces and internal line-of-business systems alike, both at production scale.
