AI systems, end to end.

Full-stack AI engineer

I design agent workflows, product interfaces, distributed services, and cloud infrastructure that survive production.

Start a conversation See selected work
8years shipping production systems
6d to 2hAI grading turnaround
93%grading accuracy across 5,000 students
1M+daily API requests supported

The model is one part of the product.

How I work

Useful AI needs product judgment, grounded retrieval, evaluation, resilient workflows, and infrastructure that can be operated after launch.

I work across that entire surface: from understanding a user's workflow and shaping the interface to designing asynchronous services, observability, deployment, and failure recovery.

Working software, not AI demos.

Selected builds

Independent projects exploring agent runtimes, retrieval, guardrails, and human-facing control surfaces.

Next.js UIFastAPIRedis queuePython workerOpenClawPostgreSQL execution + runtime events
Next.js UIFastAPI

API queues work

RedisPython worker

Shared storage & execution

PostgreSQLOpenClaw
Runtime architecture · from the README
Full-stack agent runtimeNext.js / FastAPI / PostgreSQL / Redis

Agent Orchestration Platform

Problem
Agent workflows span execution, storage, and message delivery.
Approach
Connect a visual workflow app to OpenClaw through queued workers.
Result
Inspect node state, tool calls, token usage, and estimated cost.
View project (opens in a new tab)
“revenue by region”
Live schemaValidated SQL
SELECT onlyKnown columnsRow limit
“drop orders table” → refused
Documented question & guardrails
Guardrailed analytics agentPython / OpenAI / sqlglot / Streamlit

NL-to-SQL

Problem
Natural-language questions can produce unsafe or invalid SQL.
Approach
Ground generation in the live schema and validate before execution.
Result
Read-only queries with refusal paths, row limits, and result previews.
View project (opens in a new tab)
Documents
APIConnected data
Contextual answerRetrieved knowledgeOrganization context
Document grounding · product architecture
Self-hosted agentic RAGFastAPI / Qdrant / LangChain / LlamaIndex

Vidura

Problem
Enterprise answers are buried in documents and connected systems.
Approach
Index organizational knowledge and retrieve context for each question.
Result
EAM360 field technicians get contextual answers from maintenance manuals.
View project (opens in a new tab)
Docker desktopUbuntu · noVNC
SSHMouse + keyboard
ScreenshotModel feedback
Computer-use control loop · from the README
Computer-use agentAgents / Docker / Local automation

SkyCUA

Problem
Desktop automation needs actions and visual feedback in the same loop.
Approach
Control a Docker desktop over SSH and return screenshots to the model.
Result
Local mouse, keyboard, and screenshot tools behind a web interface.
View project (opens in a new tab)
goclawCLI / HTTP
$ go run ./cmd/goclaw

/provider
/memory brief
/sandbox list
$ go run ./cmd/goclaw --http
Documented commands · terminal & service
Agent CLI and HTTP serviceGo / LLM tools / Persistent memory

GoClaw

Problem
Tool-driven tasks need context that survives across sessions.
Approach
Combine sandboxed code, model routing, and file-based memory in Go.
Result
Run the agent through a terminal UI or an HTTP service.
View project (opens in a new tab)

Systems ownership at every layer.

Experience

Remote product engineering across education, enterprise AI, payments, analytics, healthcare, and consumer platforms.

Jan 2025 to present Springboard AI grading, mentorship, workflow reliability

Software Engineer II, Full-stack

Builds AI-assisted education workflows and the distributed services behind grading, mentorship, and student support.

Reduced grading turnaround from 6 days to 2 hours at 93% accuracy across 5,000 students.

Jan 2024 to Nov 2024 TarkaLabs Self-hosted RAG, retrieval, multimodal ingestion

Technical Consultant

Delivered an enterprise agentic RAG platform spanning ingestion, hybrid retrieval, reranking, NL-to-SQL, NL-to-API, and evaluation.

Added RAGAS evaluation, PII filtering, prompt safety, output validation, and source-level traceability.

Jul 2023 to Dec 2023 Nagarro Payments, gifts, distributed transactions

Senior Software Engineer

Reworked high-volume gift and payment services using event-driven communication, Saga patterns, caching, and fault isolation.

Supported more than 1 million API requests per day while reducing cascading failure risk.

Jul 2021 to Jun 2023 Dentsu World Services Analytics, event ingestion, cloud data platforms

Software Engineer

Built Go and Python services, high-throughput event ingestion, external APIs, webhooks, and AWS processing pipelines.

Served 6,000+ daily users and more than 200,000 API requests per day.

Dec 2019 to Jul 2021 Cura Technologies Healthtech, Kubernetes, platform migration

Software Engineer

Built healthcare APIs, clinical reporting services, mobile functionality, and the platform migration to Kubernetes on AWS.

Created a Kubernetes operator for PostgreSQL provisioning, failover, and lifecycle management across 10+ hospitals.

Oct 2018 to Oct 2019 Tapchief / Relevel Product engineering, onboarding, real-time communication

Product Engineer

Shipped end-to-end features across backend APIs, Angular, React Native, WebRTC, and Twilio for web and mobile products.

Supported roughly 30,000 daily active users and 80 to 100 audio or video calls per day.

One engineer across the product boundary.

Technical range

AI systems

  • Agent workflows
  • Agentic RAG
  • Tool calling
  • Hybrid retrieval
  • RAGAS evaluation
  • Prompt safety

Product engineering

  • React
  • TypeScript
  • FastAPI
  • Django
  • Node.js
  • API design

Platforms

  • Go
  • Kubernetes
  • AWS
  • GCP
  • Terraform
  • CI/CD

Distributed data

  • PostgreSQL
  • MySQL
  • Redis
  • Qdrant
  • RabbitMQ
  • Snowflake

Need AI that works beyond the prototype?

Build the difficult part

I am interested in senior full-stack AI, backend platform, and product engineering roles where system ownership matters.

Email Ratnadeep GitHub profile Read resume