Skip to content
NJNakshatra Joshi
Available Jan 2027/Software Engineer/India

I build agents that
survive production.

I'm Nakshatra Joshi — a software engineer working where LLMs meet real backends. Typed services on PostgreSQL, Redis-backed job pipelines, and tool-calling agents that are gated by checks rather than trusted on vibes.

agent.runtimeillustrative·run 7,034
  1. receiveuser request
  2. plan—
  3. tool call—
  4. verify—
  5. respond—
running…
7,000+
PODs processed in productionCargoFL
82%
documents auto-reconciledCargoFL
10K+
users served across 2 regionsGoBrics
10
agent tools across the RFQ lifecycleCargoFL
02How I work

Four rules I've paid for.

Every one of these exists because a system taught it to me the expensive way.

01

Make correctness checkable

An LLM's output is a guess until something can verify it. Line-item reconciliation, schema validation and eval harnesses turn guesses into claims you can gate on.

02

Give the model a state machine

Agents get a tool per legal transition, not free rein over the workflow. The model decides what happens next; the system decides what is allowed.

03

Degrade, don't die

Upstream providers rate-limit and fall over. Typed error paths, retries with sweep intervals, and fallbacks keep a session alive instead of dropping it.

04

Boring deploys

Type-check and lint gates, SHA-tagged images, health-checked non-root containers. A rollback should be redeploying a known-good tag, not an incident.

04Experience

Where the work happened.

Three roles, two of which put code in front of paying customers.

May 2026 — Aug 2026

Pune, Maharashtra · Hybrid

Software Engineer Intern

CargoFLInnoctive Technologies Pvt. Ltd.

Two production systems: a chat-based freight-procurement agent and the backend of an AI document-extraction platform.

  • Built a chat-based freight-procurement agent on Fastify + React with PostgreSQL/Drizzle, implementing 10 tools for the end-to-end RFQ lifecycle across 7 carriers, plus background workers for stale-RFQ reminders and failed-job retries.
  • Built the backend of an AI proof-of-delivery extraction platform: file-type routing into a Redis queue with a Postgres-backed job table, pulling documents from S3 for an OCR + vision-LLM extraction service — 7,000+ PODs processed in production.
  • Extended line-item-to-total reconciliation across every file-type path, auto-clearing ~82% of production documents and routing the rest to manual review; implemented RBAC across 5 roles.
  • Built a write-only BYOK (bring-your-own-key) credential system: carrier API keys and secrets encrypted at rest in PostgreSQL and masked/redacted in the UI, so a stored credential can be used but never read back out.
  • Shipped CSV bulk uploads, document-to-shipment linking and URL-persisted AND/OR filters in React/TypeScript/shadcn.
  • Containerised the services and ran deploys through GitHub Actions CI/CD to AWS EC2; validated with MCP-driven Playwright E2E flows before client UAT.
FastifyTypeScriptReactshadcn/uiPostgreSQLDrizzleRedisBYOK credentialsEncryption at restRBACDockerGitHub ActionsCI/CDAWS EC2AWS S3Cloudflare WorkersOCR + vision LLMsLLM tool callingMCPPlaywright

Jun 2025 — Aug 2025

Delhi · Remote

AI Developer Intern

Prodigal AI

RAG content-generation pipelines with human-in-the-loop approval, and the MLOps plumbing to swap models without redeploying.

  • Built FastAPI backends for RAG-based AI content-generation pipelines, orchestrated as Airflow DAGs with human-in-the-loop steps that paused for reviewer approval or clarifying input.
  • Persisted conversation history in a separate store so paused runs could resume without losing context.
  • Used the MLflow Model Registry (DagsHub) to load versioned models at runtime, enabling model swaps without redeploys.
  • Containerised services with Docker Compose and deployed to AWS EC2.
FastAPIPythonRAGAirflowMLflowDagsHubDockerDocker ComposeAWS EC2

Jul 2024

On-site

Vocational Trainee — Data Visualisation

Tata Steel Ltd.

Dashboarding for sales analysis, and the data pipelines feeding it.

  • Built and deployed 3+ interactive Tableau dashboards for sales data analysis — a path to ~25% faster reporting turnaround.
  • Studied the Google Cloud Platform and SAP-BODS pipelines feeding those dashboards.
TableauSQLSAP-BODSGCP
05Stack

Tools I reach for without looking them up.

Listed by what I've shipped with, not what I've read about.

Languages
  • TypeScript
  • JavaScript
  • Python
  • Java
  • SQL
  • Bash
Backend & Web
  • Node.js
  • Fastify
  • Express
  • Hono
  • FastAPI
  • Bun
  • React
  • Next.js
  • REST
  • OAuth2 / WebAuthn
AI & Agents
  • RAG
  • Tool calling
  • MCP
  • ChromaDB
  • LangChain
  • OCR + vision LLMs
  • MLflow
  • Airflow
  • Whisper / TTS
Data, Infra & CI/CD
  • PostgreSQL
  • Redis
  • Prisma / Drizzle
  • MongoDB
  • Docker
  • Docker Compose
  • GitHub Actions
  • CI/CD pipelines
  • AWS EC2 / S3
  • Cloudflare Workers / D1
  • Yandex Cloud
  • Linux
06Contact

Hiring for backend or applied-AI work?

Available Jan 2027 — 6-month internship or full-time