Available · Remote · IST
[ Portfolio · 2026 ]
Howrah, IN
Avik Bhanja.
Full-Stack Engineer shipping production systems in NestJS, AI hybrid search (pgvector + LLM ranking), and AWS. Currently at a US-based company — migrated 87% of a monolith to NestJS 11 with zero downtime.
192
endpoints migrated
17×
TTFB speedup
43
RN screens solo
6
languages localized
[ 00 / Manifesto ]
I build production systems, not demos. Every line I ship runs in a real monolith, handles real traffic, and survives real failures. I care about TTFB, type safety, and time-to-ship — in that order.
~/manifesto.ts
01
const ship_first = "Working software beats perfect architecture. Then iterate with data.";
02
const type_safe_by_default = "TypeScript end-to-end. Prisma + Zod at the boundaries. No any, no lies.";
03
const measure_or_it_didnt_happen = "TTFB, LCP, CLS, error rate. If it isn't instrumented, it didn't improve.";
[ 01 / Selected Work ]
Things I've shipped.
Side-projects, production tools, and open-source experiments — all deployed, all clickable.
2024Live
1
2024Live
Q
2025Live
C
2026Live
N
2025Live
P
2026Live
[ 02 / Case Studies ]
The work behind the numbers.
01
Curriculo · US-Based
NestJS 11PostgreSQLpgvectorAWSCloudflare
Migrated 87% of a production monolith to NestJS 11 — zero downtime.
The problem
A 192-endpoint Express.js monolith needed to move to NestJS without breaking 30+ controllers in production.
The approach
- Dual-version routing (/v1 + /v2) running simultaneously
- Moved 49 services behind a typed DI boundary
- CI/CD on GitHub Actions — lint, type-check, health checks, cache invalidation
- DNS migration AWS Route 53 → Cloudflare, adding DDoS at zero cost
17×
TTFB speedup
48 → 85
Perf score
80 → 88
SEO score
0s
Downtime
02
Curriculo · AI Systems
pgvectorClaudeGeminiPostgreSQLNestJS
Built an AI hybrid search engine — pgvector + LLM ranking.
The problem
ATS candidate search needed to rank on meaning, not keywords — resumes, skills, roles, institutions.
The approach
- pgvector semantic embeddings + Postgres FTS in a single query
- Reciprocal Rank Fusion to combine lexical + semantic signals
- LLM-powered entity extraction: SKILL, COMPANY, ROLE, INSTITUTION
- Served inside a 33-page ATS platform with drag-and-drop Kanban + Gmail/Calendar sync
Hybrid
Search modes
4 types
Entities extracted
33
Frontend pages
Gmail + GCal
Realtime sync
03
Mahakarshan · Farmology
React Nativei18nRazorpayREST
Solo-shipped a cross-language farming app to Google Play.
The problem
Rural Indian farmers needed a production app in their language — with voice, payments, and a full e-commerce flow.
The approach
- 43 screens shipped across 18 releases (v6.1 → v7.6) solo
- i18n in EN, HI, BN, MR, TE, OD with voice input + speech-to-text
- Catalog → cart → Razorpay → order history → AI chatbot with voice
- Integrated 30+ REST APIs across product, auth, orders, content
43
Screens
18
Releases
6
Languages
30+
APIs wired
[ 03 / Experience ]
Where I've built.
Jan 2026 — Present
Curriculo
Full-Stack Developer
Remote · US-Based
- Migrated 87% of production API from Express.js to NestJS 11 — 192 REST endpoints, 30 controllers, 49 services, zero downtime via dual-version routing.
- Built AI-powered hybrid search engine combining pgvector semantic search, full-text search, LLM ranking (Reciprocal Rank Fusion), and entity extraction.
- Migrated DNS from AWS Route 53 to Cloudflare — TTFB 3.4s → <200ms (17× faster), perf 48 → 85, SEO 80 → 88, DDoS protection at zero cost.
- Architected full-stack ATS with 33 frontend pages, drag-and-drop Kanban, interview scheduling, real-time Gmail/Google Calendar sync for multi-tenant orgs.
- Built GitHub Actions CI/CD deploying 5 static sites to AWS S3 + CloudFront and NestJS API to EC2 via PM2 — lint, type-check, health checks, cache invalidation.
Aug 2025 — Jan 2026
Mahakarshan Technologies
React Native Developer (Intern)
Remote
- Solo developer for Farmology — 43 screens, 18 production releases (v6.1 → v7.6) to Google Play, 30+ REST APIs integrated.
- Implemented i18n across 6 Indian languages (EN, HI, BN, MR, TE, OD) with voice input, speech-to-text, and graceful fallbacks.
- Shipped full e-commerce: catalog, cart, Razorpay, order history, and AI chatbot with voice input and product recommendation cards.
May 2025 — Aug 2025
HomeGeeni
Full-Stack Intern
Remote
- Reduced frontend build size from 2GB+ to 144MB via code-splitting and deduplication — significantly improved Core Web Vitals.
- Built responsive UI with Next.js, TypeScript, Tailwind, and Shadcn UI using DRY component architecture.
[ 04 / Stack ]
Tools I reach for.
Not a dependency graveyard — every tool below is in something currently running in production.
01
Languages
TypeScriptJavaScriptPythonSQL
02
Frontend
Next.js 15React 19React NativeTailwindShadcn UIRedux ToolkitFramer Motion
03
Backend
NestJS 11Node.jsExpressPostgreSQLMongoDBPrismaSupabase
04
AI · Search
pgvectorClaude APIGeminiOpenAIVector EmbeddingsHybrid Search (RRF)
05
Cloud · DevOps
AWS (EC2, S3, CloudFront, Route 53, IAM)GCPCloudflareDockerVercelGitHub ActionsPM2
06
Tooling
GitClerkStripeRazorpayPostmanSwagger/OpenAPIFigmaLinux
[ 05 / Currently ]
Right now.
Howrah, India · IST
Shipping
NestJS 11 rollout + hybrid AI search at Curriculo.
Learning
Advanced pgvector, Cloudflare Workers, and production RAG patterns.
Reading
Designing Data-Intensive Applications · Martin Kleppmann.
Open to
Remote full-stack roles & select freelance (AI / NestJS / Next.js).
Let's build
Got a hard problem? Let's ship it.