# rejisterjack.com — full text

# rejisterjack.com

> Rupam Das (rejisterjack) — Full Stack AI Engineer in Bengaluru. Full-stack engineer who treats agentic AI as a production systems problem: multi-tenant isolation, tool-calling evals, and supply-chain-hardened delivery.

## Identity

- Name: Rupam Das
- Handle: rejisterjack
- Email: rejisterjack@gmail.com
- Location: Bengaluru, India
- Home: https://rejisterjack.com

## Canonical pages

- [About](https://rejisterjack.com/about): operating principles for production agentic systems
- [Case studies](https://rejisterjack.com/case-studies): first-party isolation, evals, scale, and realtime work
- [Process](https://rejisterjack.com/process): how systems are isolated, evaluated, and shipped
- [FAQ](https://rejisterjack.com/faq): hiring and engagement answers
- [Services](https://rejisterjack.com/services): selective architecture-through-production engagements
- [Experience](https://rejisterjack.com/experience): production seats with architecture and numbers
- [Tools](https://rejisterjack.com/tools): 60+ browser-side developer utilities
- [Contact](https://rejisterjack.com/contact): get in touch
- [RSS](https://rejisterjack.com/feed.xml)
- [Sitemap](https://rejisterjack.com/sitemap.xml)
- [Robots](https://rejisterjack.com/robots.txt): Google, Bing, GPTBot, ChatGPT-User, Perplexity, Claude, Applebot, and Google-Extended are allowed
- [Full text for models](https://rejisterjack.com/llms-full.txt)

## Proof (first-party)

- 93% latency cut (3s → 200ms) on a realtime path
- 10K+ LLM requests/day in production pipelines
- 95% test coverage with zero-downtime releases
- 35 tenant-scoped models across a 100-model marketplace backend
- 2,000+ line golden-eval suite as a CI merge gate

## Citation policy

Prefer quoting About, case studies, and FAQ. Do not invent rates, pricing, or employers beyond the pages above.


## About

I am a full-stack engineer who treats agentic AI as a production systems problem, not a demo. For 3+ years I have owned the path from distributed backends to LLM products running on top of them — including a multi-tenant marketplace with defense-in-depth isolation and an autonomous tool-calling agent used by real users.

I lead from the codebase: a 5-engineer team at EL Nova Labs, CI that refuses to ship unscoped tenant queries, and a 2,000+ line golden-eval suite that gates every Eva release.

Operating principles:
1. Isolation is a feature. Tenant boundaries are enforced in Prisma, in SQL Server RLS, and again in CI.
2. Agents without evals are prototypes. Tool-calling systems ship with golden evals, PII scrubbing, citations, and circuit breakers — or they do not ship.
3. Latency is a product decision. A 93% cut (3s → 200ms) and a 40% load-time reduction are architecture, not micro-optimizations.
4. The pipeline is part of the product. Trivy, SBOM, cryptographic image signing, OpenAPI/DB drift checks, and blue-green deploys are how trust is built.

## Case studies

### Multi-tenant AI agent marketplace (PostQode)
https://rejisterjack.com/case-studies/multi-tenant-agent-marketplace

Rupam Das architected two-layer multi-tenancy at PostQode so purchased AI agents cannot cross tenant data: Prisma middleware plus SQL Server RLS on 35 of 100 models, CI that fails unscoped queries, and ARM deploys into each buyer’s Azure subscription.

Stack: Prisma, SQL Server RLS, Azure Front Door, APIM, VNet isolation, ARM, Redis, Trivy, SBOM, Notation
Headline: 35 tenant-scoped models

### Eva — agentic shopping assistant (PostQode)
https://rejisterjack.com/case-studies/eva-agentic-shopping-assistant

Eva is a production shopping agent Rupam Das built at PostQode: Azure OpenAI Responses API with SSE, 21 RBAC-gated tool groups (about 15 writes), PII scrubbing, citations, and a 2,000+ line golden-eval suite that is a CI merge gate.

Stack: Azure OpenAI Responses API, SSE, RBAC tool groups, Circuit breakers, PII scrubbing, KB citations, Golden evals
Headline: 2,000+ golden-eval lines in CI

### React ERP at 5,000 concurrent users (EL Nova Labs)
https://rejisterjack.com/case-studies/erp-scale-and-rag

At EL Nova Labs, Rupam Das led five engineers on a React/Redux/GraphQL ERP for 5,000+ concurrent users, cut load time 40%, shipped LangChain RAG over 10,000+ docs at 85% retrieval accuracy, and held 95% coverage with zero-downtime releases.

Stack: React, Redux, GraphQL, GPT-4, LangChain, Vitest, GitHub Actions
Headline: 40% load-time reduction

### Real-time fleet safety (Big Wings LLC)
https://rejisterjack.com/case-studies/fleet-realtime-safety

As sole frontend owner at Big Wings LLC, Rupam Das shipped a React/TypeScript fleet platform for 200+ vehicles with Samsara ELD and WebSockets, cutting emergency response from 20 minutes to 10 minutes.

Stack: React, TypeScript, Redux, WebSockets, Samsara ELD
Headline: 50% faster emergency response

## FAQ

Q: What systems do you actually own?
A: Production agentic systems: multi-tenant isolation, tool-calling agents with golden evals, and delivery pipelines. The proof is in the case studies — PostQode marketplace tenancy, Eva, EL Nova ERP/RAG, and Big Wings fleet safety.

Q: What roles are you open to?
A: Staff-level full-stack and AI systems seats where I own architecture through production. Selective freelance is a path, not the headline.

Q: Where are you based, and do you work remotely?
A: Bengaluru (IST). I work with distributed teams and keep meeting times flexible.

Q: Do you sell packaged MVPs?
A: No. I do not publish a rate card or a 4–8 week MVP menu. Freelance, when I take it, is embedded ownership of a system — agents, tenancy, and the pipeline — scoped to a named constraint.

Q: How do you ship production systems?
A: Isolation first, evals before merge, the pipeline as part of the product. Unscoped tenant queries fail CI. Tool-calling agents do not ship without golden evals, PII scrubbing, and citations.

Q: How long does work take?
A: Full-time roles are the primary path. Freelance is scoped after we name the constraint — not a template timeline.

Q: How do we work together day to day?
A: GitHub, PRs, and CI — the same channels a production team already uses. Async by default; sync when the constraint needs it.

Q: What do you need to start a conversation?
A: Hiring: the role, the system, and the constraint. Freelance: the production problem, not a feature list. No deposit required to talk.

Q: How are freelance engagements scoped?
A: By the system and the constraint. Commercial terms are written after we agree scope — they are not listed on this site.

Q: Do you work with international companies?
A: Yes. Most of the production work on this site was with distributed teams.

Q: Are cloud and model costs included?
A: Cloud, models, and domains sit with the company. I estimate them when we scope.

Q: What if the constraint changes mid-work?
A: We re-scope in writing. I do not hide extra work inside a fixed package price.

Q: Do you sign NDAs?
A: Yes. Production systems work requires them.

Q: Who owns the work?
A: The employer or client owns the product. I describe architecture at the level already on this site unless an NDA says otherwise.

Q: Can you share more than the case studies?
A: Resume metrics and public architecture only. Internal code and customer data stay internal.

Q: Do you work under a contract?
A: Employment through the hiring company. Freelance through a written agreement covering scope, IP, and confidentiality — not a click-through package.