Scene 02 · Take 01 — New York, NY

BIMAL
KUMAL

> AI agent engineer

40.7128°N · 74.0060°W — shot on location in New York
github.com/bimal1023 linkedin.com/in/bimal-kumal
keep rolling
Reel 01

DAILIES

selected engineering work, straight off the light table

Cel · 001

ARTHVION

Multi-Agent PE Diligence Platform · Founder

2026 — Present

TypeScriptNext.jsPython FastAPICeleryRedis Claude · BedrockMCPpgvectorStripe
  • 4–5 days of analyst work compressed into a 4-minute, fully-cited investment memo — four specialist agents (Financial, Risk, Market, Legal) read SEC filings, the web, and private deal-room docs in parallel.
  • ~90% lower inference cost and 5x faster after profiling the agent pipeline against a fixed eval workload.
  • Shipped the full multi-tenant SaaS solo: real-time streaming UI over SSE, Stripe billing, an atomic usage ledger, and Terraform deploys.
↗ GitHub
hover to bring the blueprint to ink
Cel · 002

SENSEI

Real-Time Voice Agent

Sept 2026 — Present

TypeScriptReactWebRTCFastAPIOpenAI Realtime
  • Sub-second voice responses with interruption handling via the OpenAI Realtime API — speech-to-speech, tool calls, and long-term memory.
  • Zero API keys exposed to the browser: short-lived Realtime session keys are minted server-side, with live tool calls routed through FastAPI.
↗ GitHub
hover for the call flow
Cel · 003

STELLANET

Researcher Matching & Ranking Platform

March 2026

Amazon Bedrock · NovaPythonFastAPIReactOpenAlex
  • Top 100 of 13,000+ entrants in the AWS Nova AI Hackathon with an LLM system ranking researchers across OpenAlex.
  • Zero hallucinated output fields by enforcing a strict JSON schema contract on every LLM response; 4x lower latency via parallel fan-out and caching.
↗ Live Site ↗ GitHub
hover for the ranking pipeline
Cel · 004

PULSE

Personal AI Assistant & Automation Agent

2026

FastAPIPythonGPT-4.1 miniReactAPSchedulerDiscord
  • Autonomous tool-calling agent with 11 tools — web search, news, Arxiv, GitHub trends — streaming answers over SSE in a ChatGPT-style interface.
  • 10 scheduled pipelines a day: job matching scored against my resume, daily DSA/system-design drills, and research digests, delivered over Discord and email.
↗ Live Site ↗ GitHub
hover for the agent loop
Cel · 005

PILOTPHD

AI PhD Application Co-Pilot

2026

Next.jsReactTypeScriptFastAPIPostgreSQLOpenAlex
  • Email drafter, personal-statement refiner, fellowship finder, and professor finder — four AI agents in one PhD-application workspace.
  • Faculty search ranks professors by fit across OpenAlex — the same engine behind Stellanet, repointed at a different audience.
↗ Live Site ↗ GitHub
hover for the agent set
Reel 02

TOOL DESK

what's in the drawer

AI Agents
  • Tool Use
  • MCP
  • Multi-Agent Orchestration
  • Voice Agents
  • Realtime API
  • WebRTC
  • Prompt Engineering
Evals & Retrieval
  • LLM Eval Frameworks
  • Regression Evals
  • RAG
  • Hybrid Retrieval
  • pgvector
  • Full-Text Search
Languages
  • Python
  • TypeScript
  • JavaScript
  • SQL
  • Java
Full-Stack & Infra
  • React
  • Next.js
  • FastAPI
  • PostgreSQL
  • Redis
  • Docker
  • Kubernetes
  • Terraform
  • CI / CD
Reel 03

MODEL SHEET

character notes & production history

3.95
GPA
700+
Employees Reached
91.7%
RAG Recall@5
$115K/mo
Cloud Savings Surfaced

I'm an AI agent engineer who ships to production — not demos. Three agents live for ~700 employees at TransRe, a RAG agent holding 91.7% recall@5, and a real-time voice agent that talks back in under a second. I founded Arthvion, a multi-agent diligence platform, from a blank repo to a billed SaaS. Python, TypeScript, React — one frame at a time.

Experience
Agentic AI Engineer Intern June – Aug 2026

TransRe · New York, NY

3 production AI agents shipped to ~700 employees in 10 weeks as sole engineer. 91.7% recall@5 conversational RAG agent via hybrid pgvector + full-text retrieval. $115K/month in cloud savings surfaced with a 50+-tool MCP agent across 6 billing platforms.

Education
St. Joseph's University, New York Expected May 2027

BS in Computer Science & Data Science · GPA 3.95/4.00

Honors Thesis: LLM-Based Multi-Agent Workflows for Automated Risk Intelligence · AWS AI/ML Scholar · CodePath Applied AI Engineering Fellow · Dean's List every semester

Reel 04

APPROVAL REEL

stamped, signed, cleared for release

Approved
Merit · Leadership
ESSE NON VIDERI SCHOLAR

90% tuition scholarship awarded for outstanding academic merit and demonstrated leadership.

Top 100
AWS Nova AI Hackathon · 13,000+ entrants
STELLANET: TOP 100 FINISH

Ranked in the top 100 of 13,000+ entrants with an LLM system ranking researchers across OpenAlex.

Certified
Accenture
GENERATIVE AI WITH AWS

Certification in generative AI fundamentals and AWS integration.

Completed
Stanford Online · Andrew Ng
SUPERVISED MACHINE LEARNING

Completed Stanford's ML specialisation covering regression, classification, and neural networks.

ROLL
CREDITS.
bimalkumal2004@gmail.com 917-821-6444 linkedin.com/in/bimal-kumal github.com/bimal1023