Lanesurf Verified
["Sports & Recreation","Leisure","Engineering","Manufacturing"]
Principal ML Research Engineer
San Francisco, California, United StatesOnsiteFull TimePrincipal$120,000–$225,000 /yrPosted 2 months agoHidden Gem · YC Startup
Role summary
We are seeking a Principal ML Research Engineer to build AI agents that automate the process of booking truck shipments. This role involves fine-tuning speech and transcription models for real-time voice agents, optimizing LLMs for efficiency and accuracy in complex conversations, and defining evaluation metrics for model performance. The ideal candidate has experience training and deploying models in production, is proficient in Python and PyTorch, and thrives in a fast-paced startup environment. This is an onsite role in San Francisco, focused on solving hard problems in autonomous logistics and real-world transactions.
# Build AI that talks, negotiates rates, and enables autonomous movement of trucks from pickup to delivery
### [**Demo of AI booking a shipment in 10 minutes by speaking to 96 trucking companies simultaneously**](https://www.linkedin.com/feed/update/urn:li:activity:7394069447327555584)
### The problem
If **Walmart** needs to move a truck of avocados from **California to Chicago**, today they must:
* Speak with **50+ trucking companies**
* Check **weight and temperature requirements**
* Negotiate **price and availability**
* Do it **one call at a time**
This process takes hours and thousands of phone calls every day across the industry.
### What we’re building
We’re building **AI agents** that do this work **automatically**.
* Calls and emails **dozens of trucking companies at once**
* Checks requirements (weight, temperature, lanes)
* **Negotiates prices in parallel**
* Books a truck **in minutes, not hours**
### Proof it works
👉 In this demo, our AI **spoke to 96 trucking companies simultaneously and booked a shipment in under 10 minutes -** https://www.linkedin.com/feed/update/urn:li:activity:7394069447327555584
### Why this is exciting
* You’ll work on **AI that handles real-world transactions through phone calls**
* Real-world, high-stakes work enabling autonomous logistics - think moving a truck from Chicago to Texas, fully coordinated by AI
* Small team, high ownership, fast iteration
* Hard problems that don’t exist in benchmarks
# What we’ll work on
### Train & Tune Models
Fine-tune **transcribers and speech models** for real-time voice agents operating on live phone calls.
* Enable **real time** **transcriber** fine-tuning based on caller context
* Improve **transcription accuracy** for **domain-specific language** under noisy conditions
* Fine-tune **interruption models** on domain-specific conversations
* Post-Train **speech models** for intonations, pacing and naturalness and avoiding robotic cadence
### LLM optimization
* Structuring modules, and policies that compose cleanly
* Optimizing LLM outputs for **brevity, correctness, and timing**
* Reducing drift across long, multi-turn conversations
* Evaluating changes against **real call outcomes**, not just text metrics
### Evaluation & iteration
You’ll help define how we measure quality across:
* Transcription accuracy where it actually matters
* Voice naturalness as judged by listeners
* Conversation efficiency and completion
# You can be a great fit, if:
* **ML Engineer with Real-World Experience** – You’ve trained and shipped models in production. Bonus if you’ve worked with LLMs or audio models.
* **Fluent in Modern ML Stack** – You know your way around Python, PyTorch, and today’s ML tools - from training pipelines to evaluation benchmarks.
* **Execution-Oriented** – You move fast, take ownership, and focus on solving real problems over perfect ones.
* **Startup-Ready** – You’re adaptable, resilient, and energized by ambiguity and fast-changing priorities.
* **Clear Communicator & Team Player** – You collaborate well across functions and push decisions forward.
# Details
* **Cash + Equity**
* **Location**: San Francisco, CA, US
### [**Demo of AI booking a shipment in 10 minutes by speaking to 96 trucking companies simultaneously**](https://www.linkedin.com/feed/update/urn:li:activity:7394069447327555584)
### The problem
If **Walmart** needs to move a truck of avocados from **California to Chicago**, today they must:
* Speak with **50+ trucking companies**
* Check **weight and temperature requirements**
* Negotiate **price and availability**
* Do it **one call at a time**
This process takes hours and thousands of phone calls every day across the industry.
### What we’re building
We’re building **AI agents** that do this work **automatically**.
* Calls and emails **dozens of trucking companies at once**
* Checks requirements (weight, temperature, lanes)
* **Negotiates prices in parallel**
* Books a truck **in minutes, not hours**
### Proof it works
👉 In this demo, our AI **spoke to 96 trucking companies simultaneously and booked a shipment in under 10 minutes -** https://www.linkedin.com/feed/update/urn:li:activity:7394069447327555584
### Why this is exciting
* You’ll work on **AI that handles real-world transactions through phone calls**
* Real-world, high-stakes work enabling autonomous logistics - think moving a truck from Chicago to Texas, fully coordinated by AI
* Small team, high ownership, fast iteration
* Hard problems that don’t exist in benchmarks
# What we’ll work on
### Train & Tune Models
Fine-tune **transcribers and speech models** for real-time voice agents operating on live phone calls.
* Enable **real time** **transcriber** fine-tuning based on caller context
* Improve **transcription accuracy** for **domain-specific language** under noisy conditions
* Fine-tune **interruption models** on domain-specific conversations
* Post-Train **speech models** for intonations, pacing and naturalness and avoiding robotic cadence
### LLM optimization
* Structuring modules, and policies that compose cleanly
* Optimizing LLM outputs for **brevity, correctness, and timing**
* Reducing drift across long, multi-turn conversations
* Evaluating changes against **real call outcomes**, not just text metrics
### Evaluation & iteration
You’ll help define how we measure quality across:
* Transcription accuracy where it actually matters
* Voice naturalness as judged by listeners
* Conversation efficiency and completion
# You can be a great fit, if:
* **ML Engineer with Real-World Experience** – You’ve trained and shipped models in production. Bonus if you’ve worked with LLMs or audio models.
* **Fluent in Modern ML Stack** – You know your way around Python, PyTorch, and today’s ML tools - from training pipelines to evaluation benchmarks.
* **Execution-Oriented** – You move fast, take ownership, and focus on solving real problems over perfect ones.
* **Startup-Ready** – You’re adaptable, resilient, and energized by ambiguity and fast-changing priorities.
* **Clear Communicator & Team Player** – You collaborate well across functions and push decisions forward.
# Details
* **Cash + Equity**
* **Location**: San Francisco, CA, US
