Product Engineer, New Graduate

What is Nooks?

About Nooks

• The team: Nooks is ~45 people. Engineering & product are mostly in SF and go to our office 2-3x/week. The go-to-market team is distributed across the U.S.
• The founders (Dan, Rohan, and Nikhil) met studying AI at Stanford, have published in top AI journals, Forbes 30u30, worked at Scale AI, Tesla Autopilot, etc.
• The engineering team has won international math & physics olympiads, has experience at Google, Facebook, Slack, Quora, Scale AI, Bolt, Snap, Flexport, and other fast-growing startups.
• The sales team have been top-performers at companies like Gong, Amplitude, LeadIQ, and Orum.
• Fast growth: We’ve grown $0 → ~$4M ARR in 20 months. We grew 4x in 2023 and expect to 3x by EOY 2024.

The problem

Most of the sales rep’s job can be automated with today’s technology: large language models, web scraping, automation, integrations, etc.

Nooks today

• AI dialer - automates the manual parts of the calling process: skipping answering machines, leaving voicemails, taking notes, logging calls, even figuring out what to say on a call
• Analytics - we record, transcribe, and analyze every call. Since these are all outbound calls with little context, these calls follow similar structure - opener, pitch, questions/objections, ask for meeting, etc. So we can answer questions like: “which reps struggle to book the meeting with prospects who showed interest” or “what are the most common objections across each of our key personas”
• Salesfloor - sales reps & managers can work together throughout the day, listen to each others’ calls, give real-time advice, coaching, shadowing, onboarding, training.

The role

Examples of engineering problems we’re working on

• Concurrency & distributed systems
• Our smart dialer places calls in parallel and runs a realtime AI model on each call. There are some interesting concurrency problems syncing state between Twilio, our backend, and the frontend, and knowing which calls to connect, which to continue in the background, and when to start the next call.
• Realtime audio AI & precision/recall/latency tradeoffs (algorithms & models)
• We use audio data, transcription, silence detection, and several other signals to detect whether a live phone call is a voicemail, a human, or a dial tree. Here, latency is a third factor added to the standard precision/recall tradeoff because it’s important we can detect humans quickly. Our approach involves LLM embeddings, few-shot learning, data labeling, and continuous monitoring of model performance in prod.
• Latency (infrastructure)
• If our model took 5 seconds to detect a human on a phone call, the human would hang up. It’s imperative we can detect quickly and that our users can execute calls quickly. There’s latency across the detection pipeline including transcription models, audio models, websockets, Twilio API, database transactions, etc.
• Smart call funnels & playbooks (data wrangling, backend eng, GPT-3, UX)
• At what point in the conversation do my reps get stuck? What are the toughest questions that we need to address? Can I “program” a playbook so that Nooks will help my team standardize toward best-practices? We’re using GPT-3 and other LLM’s to turn companies’ mostly unstructured call data into actionable strategies & feedback loops.
• Conversation embeddings & markov models (ML modeling)
• What does the anatomy of a call look like? If I say XYZ, what are the different ways the prospect might answer and the probabilities of each? Conditioned on the first half of the call, what do I say next to maximize the likelihood that I book a demo at the end of the call? Can we use LLM’s to generate embeddings of conversations that we can use to cluster similar conversation patterns and predict where the conversation is headed?
• Integrations
• Our dialer integrates with customers’ sales engagement platforms. Every new platform we integrate with, that opens up a larger market for our product. When building integrations, we need to make sure they’re robust, reliable, and well-abstracted.
• Frontend performance
• There’s a lot going on in the frontend - WebRTC, Twilio, React rendering, websockets, etc. And people use Nooks throughout the workday, so we need to make sure our app is performant across a wide range of devices

Requirements

• 
  Ability to work with our stack:
• Some practical software engineering experience garnered from industry internships
• Ability to work with our tech stack:
• Frontend: React, Typescript, MobX
• Backend: Node.js, Express, Typescript
• Technologies: Firebase, Firestore, Websockets, Twilio, WebRTC, Postgres, Redis
• ML: GPT, Transformers, PyTorch, signal processing, few-shot classification.
• 
  Experience building complex systems (ideally somewhat related to ours)
• 
  You’re a confident, independent, and experienced engineer who is used to extreme ownership and solving hard problems