AS

I build production AI assistants and the systems behind them.

AI engineer in New Delhi — RAG chat assistants, LLM integrations and the data pipelines underneath, on thirteen years of full stack work. 11,893 hours across 123 clients, Top Rated on Upwork with 100% Job Success.

Python/FastAPI/Claude & OpenAI APIs/PostgreSQL — since 2012

Available for new work — Oct 2026

Start a project View Upwork profile

About

Ten thousand hours,
now with a model in the loop.

Abhishek Sinha

I’m Abhishek Sinha, an AI engineer based in New Delhi. I started freelancing as a full stack developer in 2012 and never really stopped: 11,893 hours, 155 jobs, 123 clients. For the last year the work has been LLM applications, and the two systems below are live in production.

The brief is usually the same. A business already has a platform, a membership site, a CRM, a pile of transcripts and PDFs, and wants an assistant that answers from its own content without making things up. I build the retrieval, the guardrails, the admin tooling and the integration — Python and FastAPI on PostgreSQL and AWS, with Claude and OpenAI behind it.

Thirteen years on Shopify, Uscreen, WordPress and HubSpot mean I know where the data actually lives and how to wire a model into it. Most of the risk in an LLM project sits there, in authentication, data isolation, rate limits and handover, not in the model call.

Based
New Delhi, IN
Working
Remote, worldwide
Freelancing
Since 2012
Languages
English, Hindi

Selected work

Two systems,
live in production.

TheoChat AI chat interface: a grounded answer with its source document, and a compliance notice refusing a request for trade signals

TheoChat AI

Compliance-guarded RAG chat assistant for course members · TheoTrade

A multi-tenant assistant embedded in TheoTrade’s WisePub member pages. Members ask about session transcripts, trade logs and onboarding emails; every answer is grounded in that product’s knowledge base, cites the document it came from, and refuses when the material is silent.

Seven guardrail layers sit around the model — prompt-injection detection, anti-probing limits, pre- and post-LLM checks and a kill switch — validated with a test-prompt suite. FastAPI, PostgreSQL with pgvector, Celery, OpenAI and Claude, a React admin dashboard, Docker on AWS. Shipped solo as an 8-week fixed-price MVP.

RAGGuardrailsFastAPIpgvectorOpenAI · ClaudeAWS

Case write-up on Upwork
Architecture diagram of the TheoTrade warehouse: five data sources feeding bronze, silver and gold PostgreSQL layers, identity resolution, and a Next.js dashboard

Warehouse

Five platforms, one source of truth · TheoTrade

A data warehouse and analytics dashboard that pulls Keap, SamCart, Beehiiv, OptiPub and WisePub into PostgreSQL on AWS RDS in bronze, silver and gold layers, resolves one identity per person across roughly 389,000 records, and serves thirteen dashboard views in Next.js.

The incremental pipeline cut the daily run from 39 minutes to under three, and revenue reconciles to within 1% of the source systems. FastAPI, dbt, Dagster, a Lambda webhook receiver and a member portal with SSO, all documented for handover.

Data engineeringPostgreSQLdbtDagsterNext.jsAWS

Case write-up on Upwork

Certifications

Credentials, dated.

Stack

The tools, honestly listed.

AI & LLM

PythonFastAPIClaude APIOpenAI APIRAGpgvectorGuardrails & evalsClaude Code

Data

PostgreSQLRedisCelerydbtDagsterAlembic

Platforms

ShopifyUscreenWordPressHubSpot

Web

PHPLiquidJavaScriptReactNext.jsREST APIs

Tools & infra

AWSDockerNginxGitHubFigma

By the numbers

11,893

Hours worked

155

Jobs

123

Clients

100%

Job success

Services

What I actually do.

  1. AI assistants & RAG chatbots

    Chat assistants that answer only from your own content, with citations and refusals built in — retrieval, embeddings, prompt design, streaming chat UI and the admin tooling to run it.

    RAG / LLM

  2. LLM integration

    Adding Claude or OpenAI to the platform you already run — membership sites, CRMs, dashboards, back offices — with auth, multi-tenant data isolation, rate limits and cost tracking.

    Claude / OpenAI

  3. Guardrails & evaluation

    Input and output guardrails, prompt-injection defence, compliance rules and test-prompt suites, so the assistant stays inside the lines you draw.

    Safety

  4. Data pipelines & warehouses

    Connectors, medallion warehouses, identity resolution and dashboards on PostgreSQL, dbt and Dagster — the groundwork an AI feature needs before it can be trusted.

    FastAPI / Postgres

  5. Platform development

    Shopify, Uscreen, WordPress and HubSpot — custom themes, templates and plugins, built to the brief rather than bent from a template.

    Shopify / Uscreen / WP

  6. REST API integrations

    Endpoints, webhooks and third-party integrations — connecting the pieces of a stack that don’t talk to each other.

    REST / Webhooks

Background

The paper trail.

2026

Anthropic, Claude Academy

Building with the Claude API · Model Context Protocol: Advanced Topics

Certifications

2026

OpenAI Academy

Design and Build Agentic Systems

Course

2025–present

TheoTrade

AI engineer: TheoChat AI and the reporting warehouse

Contract

2018–present

Uscreen.tv

Recommended Uscreen Developer

Certification

2013–2015

CreativeHaus

Senior WordPress Developer

Role

2012–2015

Udemy

iOS Development

Self-directed

2012–2013

Studio98

Senior WordPress Developer

Role

2007–2011

Dr. M.G.R. University

B.Tech, Computer Science

On campus

Testimonials

In their words.

“Abhishek is always a fantastic communicator, meets deadlines, and is very easy to work with. I’ve had zero issues and plan on continuing to work with him as our main Uscreen developer for this project!”

GerrySadhana Live

“A very friendly and welcoming person. Good work.”

ShahidSqadia

“Great experience working on UScreen website customization. Quick turnaround, very easy to work with and meet my needs.”

Frankie TaylorRetroglow Studio

“Abhishek did a fantastic job on this website. He was able to fix any issues that I had and I was impressed at his ability to build a custom admin system for me. Will be using again.”

Victoria PaulCreativeHaus

Contact

Let’s build the thing.

Tell me what you’re building, what it has to answer from, and when you need it live.