Arpit Singhal

I build the parts underneath the feature.

Founding engineer at TwinMind — AI memory for 400k+ people. Before that: 18 enterprise systems in sync and a medical information platform in regulated pharma.

How I work

Nobody asks for APIs, state, sync, or release quality. Everybody notices the day they're wrong.

I do my best work on small teams where the problem is real and the requirements aren't settled. Lately that's meant memory systems — products that shouldn't meet you for the first time every time.

Users should stay in control of what a product remembers, corrects, or forgets. That boundary between product decisions and backend systems is the work I enjoy.

People on the product
400k+
Lower transcription cost
80%
Fewer avoidable chat failures
30%Estimated
Systems kept in sync
18

Experience

Jul 2019 — Present

Explore each role

Building AI memory infrastructure for a product used by more than 400,000 professionals and students.

  • Built chat infrastructure connecting five-plus context sources, including recordings, previous chats, Gmail, notes, and summaries.
  • Filtered silence and non-speech audio with voice activity detection before transcription, reducing STT-related LLM cost by 80%.
  • Added rate limiting, request controls, and response-flow handling on high-usage chat paths, reducing avoidable failures by an estimated 30%.
  • Shipped subscriptions, paywall experiments, and product event tracking across the application.

Studied

Illinois Tech

Jan 2023 — Dec 2024

Master of Science, Computer Science · 4.00 GPA

Selected work

01 — 04

Flagship project · 01

Half-Life

A work-week analysis tool that evaluates AI exposure task by task, then creates a practical handoff plan.

Half-Life technologies: Node.js · Anthropic SDK · SSE · Structured output

Flagship project · 02

PaperTrail

A research workspace that combines PubMed, arXiv, and uploaded documents to stream answers with source citations.

PaperTrail technologies: Next.js · PostgreSQL · Qdrant / pgvector · Redis

More projects

03 — 04

03

MailIQ

A Gmail workspace for categorizing messages and handling bulk actions, with a three-panel inbox and real-time sync.

MailIQ technologies: Remix · Gmail API · Neon Postgres · Gemini

View MailIQ source
04

Whispr

A weekend team prototype that turns live meeting transcripts into summaries, action items, and a memory bank you can question afterwards.

Whispr technologies: Remix · Firebase · Google Generative AI

Stack

TypeScript · Python · Node.js · React · PostgreSQL · Redis · AWS · Docker · pgvector

Off the clock

Learning pickleball. Still trying to survive in the swimming pool.

Contact

Open to work

Looking for founding, product, and forward-deployed engineering roles where backend systems, AI workflows, and reliability matter.

Let's talk.