Applied AI Engineer · Chennai → anywhere

I build AIsystems that ship.

Eight open-source agentic systems — RAG with CI-gated evals, a 14-tool MCP platform, multi-model gateways, a QLoRA fine-tune. By day I run analytics at Teleparty on a ~50B-event warehouse. I also trained as a Shakespearean actor, which is why my AI film studio exists — and why my demos land.

MCP platformsAgentic workflowsLLM evaluation & RAGData @ TelepartyForward-deployed
GitHub ↗LinkedIn ↗Résumé →Email ↗

I trained as a Shakespearean actor, then built the AI pipeline that automates my old craft. That's how I approach every domain — learn it deeply, then ship the system. I build AI end-to-end: interface, backend, and the model logic underneath.

Rishabh Jain
8open-source AI systems shipped
14MCP tools, green CI
13agent LLM pipeline in production
$67k/qtr bug in a ~50B-event warehouse
01

AI &
Engineering

Agentic systems, MCP tooling, full-stack delivery
MCP Platform

studio-mcp

14-tool MCP platform running a generative-film pipeline end-to-end: shot planning, stills, QC gates for style drift, animation, edit. Next.js console with schema-driven tool invocation and run history.

Case study → · Repo ↗ · Live demo ↗
MCP · Agent SDK · Next.js
Retrieval

creative-rag

Hybrid retrieval (Chroma + BM25, cross-encoder reranking) over a 470-chunk film-craft corpus. CI gate fails the build on citation-unsupported claims. FastAPI, Docker, Terraform.

Case study → · Repo ↗
RAG · evaluation · CI gates
Multi-model gateway

fusion

Fans one prompt across a frontier-model panel and synthesizes a single answer; OpenAI-compatible API on Cloudflare Workers. Published eval is honest about where it wins (cost) and where solo models win.

Repo ↗ · Live API ↗
orchestration · Workers · evals
Fine-tuning

prompt-lora

QLoRA fine-tune teaching Mistral-7B to turn short briefs into structured cinematic prompts. Claude-distilled dataset, measured before and after: structural validity 0→90%, LLM-judge score 1.30→4.55 (3.5×). 4-bit NF4 on a single T4.

Repo ↗
fine-tuning · LoRA / QLoRA · eval
Agent orchestration

langgraph-studio-demo

The studio-mcp orchestrator re-expressed as a LangGraph state machine — typed state, conditional edges, interrupt/resume human gates, checkpointing. README weighs hand-rolled control flow against the framework, honestly.

Repo ↗
LangGraph · agents · FastAPI
Knowledge systems

knowledge-os + graphify

200-page schema-enforced LLM wiki plus a tool that turns code, docs and media into a queryable knowledge graph with provenance.

knowledge graphs · retrieval · Python
AI Operating System

AIOS

A personal operating system of Model Context Protocol servers and skills — agents that call APIs, read databases, and act across connected tools in multi-step chains.

MCP · Claude Code · agents
Agentic finance

Investor OS

An agentic system pulling live market and portfolio data through MCP and structured tool calls to run automated analysis and reporting, with an append-only ledger and LLM advisor gate before any trade recommendation.

agents · live data · MCP
California Institute of Technology (Caltech)AI/ML certificate (2024–25): applied projects in classification, CV, recommenders, clustering, forecasting — on GitHub.Foundations ↗
Santa Clara UniversityB.S. Commerce, Management Information Systems (2019–24) · GPA 3.65 · minors in Business Analytics, Real Estate, Theatre.Foundations
02

The Classical AI

AI film studio · run on my own MCP pipeline

I trained as a Shakespearean actor, then built the AI pipeline that automates my old craft — an actor's eye on a new medium. Every spot below is generated end-to-end through studio-mcp: shot planning, stills, a vision QC gate, animation, cut.

LEGO World Cup — AI spot · FIFA 26, Los Angeles
Tom Ford — spec fragrance ad · generative B&W, directed pipeline
The bridge

Why an actor ships better AI

Domain expertise is the forward-deployed edge. I know performance and cinema cold, so I built production tooling for it — the same way I'd embed in any customer's domain and ship the system.

directing · taste · human-in-the-loop
AI × film — visual dev

Screenplay Mood Boards

AI-driven mood boards that visualize a screenplay — generative imagery turning script into look and tone.

generative AI · visual dev
03

Experience

Data · production · operations · risk
TelepartyData Analyst, Analytics Engineering
2026–present · RemoteRebuilt the trial-to-paid conversion metric, catching a 2.5× overstatement. Re-mapped marketing attribution (+9pt coverage), surfacing a $67k/quarter attribution bug. Built a 13-agent LLM analytics pipeline — plan, build, validate, review, publish — wired to Linear via MCP.
90 Pictures ProductionsAssistant Director
2023 · ChennaiAD on two features; generated 50+ gen-AI storyboard concepts for director ideation; coordinated a 5-person AD team.
Shuchi Beverages Pvt LtdFinancial Operations Analyst
2022–23 · ChennaiAnalyzed ~$22M in claims across 11 lender banks through tribunal proceedings.
Ernst & YoungTechnology Risk Analyst Intern
2022 · San FranciscoSOX IT general controls for enterprise clients.
04

Capabilities

What I bring to a forward-deployed team
MCP & Agent Orchestration
  • MCP server design — 14-tool platform, tool contracts & schemas
  • Anthropic Agent SDK — multi-agent orchestration
  • 13-agent production pipeline (MCP → Linear)
  • Human-in-the-loop approval gates
Evaluation & Observability
  • Langfuse tracing across multi-agent runs
  • Benchmark suites — latency, accuracy, context fidelity
  • Citation-verification (LLM entailment) → CI gate
  • Style-drift QC gates on generative output
  • Regression gating — green CI across repos
Retrieval
  • Hybrid — Chroma + BM25, reciprocal-rank fusion
  • Cross-encoder reranking
  • Grounded generation, API-key auth
Production & Infrastructure
  • Docker · Terraform / ECS Fargate
  • Cloudflare Workers — edge gateway
  • GitHub Actions CI/CD gates
  • FastAPI · REST
Languages & Frameworks
  • Python · TypeScript · SQL
  • Next.js / React
  • PyTorch
Data & Analytics
  • Athena / Presto · BigQuery
  • Superset dashboards
  • PostHog product analytics
  • Revenue & funnel instrumentation
05

Now

Data Analyst at Teleparty — analytics engineering on a ~50B events/month warehouse; conversion, attribution, growth.

Building open-source MCP tooling and an AI film studio — proof I take a domain and ship the system for it.

Next: applied-AI / forward-deployed / solutions-engineering roles — open to relocation, anywhere.

Résumé →GitHub ↗
—

Offstage

Before the systems, the stage. I toured Shakespeare across four countries and still act — it's where the taste and the directing instinct come from.

  • Romeo · Orsino · EdgarShakespeare leads · Bremer Shakespeare Co., European tour
  • Assistant DirectorArulvaan — feature, Amazon Prime 2026
  • Maxwell (lead)Glean — short film, Santa Clara
Full stage & screen résumé →