Applied AI Research and Engineering
- Salary
- $200K - $300K
- Level
- Mid
- Work type
- Hybrid · New York
Not stated in the posting: visa.
Open. First seen 2 October 2026.
About the role
About OuterSignal
We are building the customer intelligence and personalization layer for consumer businesses. Every business knows when an order comes in or a user signs up, but they only see 5% of the real story.
We show them the other 95%: not just who’s buying, but who that person actually is. That includes the execs, influencers, journalists, retail buyers, investors, and everyday customers who become a brand’s best evangelists. Our platform enriches every customer in real time with professional and personal signals, builds personas, and powers everything from surprise-and-delight outreach to smarter email flows, analytics, and BD leads.
We’re a small, high-performing team building the category-defining platform for customer intelligence, backed by a world-class investor base.
About the Role
This is an applied research role for an engineer who can train and evaluate models. You will own the full loop: turn production traces into datasets, build the evals and verifiers that define "good", post-train open models (SFT, preference optimization, RL), and serve them in production at a fraction of the cost and latency of frontier models. Depending on your background, you'll join as a Research Engineer or Research Scientist; title and level follow your skillset and experience.
What You'll Work On
- Distill multi-step, frontier-model enrichment pipelines into fine-tuned models, and prove with evals that cost, quality, and throughput improve.
- Design verifiers, graders, and environments for our multi-step enrichment agents, where the answer is only partly checkable, then find out how a model games them.
- Run post-training (DPO, GRPO) on open models against those verifiers and environments.
- Build our eval suite from error analysis of real production traces: extraction accuracy, data quality, and end-to-end customer impact.
- Ship the models you train: LoRA adapters trained with managed post-training APIs (like Tinker) and served by inference providers, routed alongside the frontier APIs they replace, with a clear read on cost, latency, and quality.
- Generate synthetic training data, and measure whether it actually helps.
- Ship all of it into our production pipelines (Temporal, Postgres, ClickHouse, Kubernetes) with observability on cost, latency, and quality.
Who You Are
- A strong software engineer: you write production code, reason through distributed systems and data pipelines, and simplify aggressively.
- You have hands-on deep learning experience: you've trained or fine-tuned models yourself, at work or in your own projects, and can tell us what broke and why.
- You hunt for failure modes. When a metric goes up, your first question is how the model cheated: reward hacking, leaked labels, a grader that rewards format over substance. You reach for ablations, controls, and held-out evals before you believe a result.
- You have taste: you know when to train, and when a better prompt, a better eval, or a smaller model is the real answer.
- Strong signals: post-training: SFT, DPO, and especially RL with verifiers, full or LoRA; reward and verifier design, environments; open-source tools or reproductions others use; public write-ups with ablations. Publications are welcome but not required.
- Bonus: deep inference engineering (vLLM/SGLang, quantization, speculative decoding) for when we bring serving in-house.
To apply: send links to one or two things you built (a repo, a reproduction, a write-up, a paper) and tell us about the most interesting failure you caught in your own work.
Why Join Now
- Real compute for research: a dedicated, serious budget for training and research experiments.
- We publish: we write up our research and share it through papers and our engineering blog.
- Early seat: you'll help shape the DNA of OuterSignal's applied research org from the beginning.
- Massive upside: meaningful equity in a venture-backed company defining a new category in commerce intelligence.
- Right moment: we're past prototype, real brands already rely on us, and we have strong product-market fit, but the ceiling is still wide open.
How to Apply
Submit your resume along with responses to the application questions. We review applications on a rolling basis and will reach out to qualified candidates to discuss potential opportunities.
OuterSignal is an equal opportunity employer. We celebrate diversity and are committed to creating an inclusive environment for all employees.
The team: Engineering HIRES STEADILY
This role is one of 3 open in Engineering. 1 was posted in the last 30 days and the oldest in December: a team that hires all year rather than one being built.
WHAT THE TEAM WORKS WITH, ACROSS ITS POSTINGS
ClickHouse (2 of 3), PostgreSQL (2 of 3)
OTHER ROLES ON THIS TEAM
- Design and Product Engineering posted Nov 16
- Backend and Infrastructure Engineering posted Dec 5
From the team and dates on OuterSignal's own job board.
Similar jobs
- Intelligent Automation/AI-ML Engineer at Node.Digital
Shares 6 skills: Hugging Face, MLflow, PyTorch… · same seniority · same country - Full-Stack AI Engineer at Pavago
Shares 6 skills: Hugging Face, MLflow, PyTorch… · same seniority · same country - Senior Product Manager, Applied AI & Machine Learning - Operator Suite at JustPark
Shares 5 skills: Jupyter, MLflow, PyTorch… · same country - Senior AI Platform Engineer at Code Metal
Shares 6 skills: Hugging Face, MLflow, PyTorch… · same country - AI Solutions Architect at Accenture Federal Services
Shares 6 skills: Hugging Face, MLflow, PyTorch… · same country - Design and Product Engineering at OuterSignal
Also at this company
More jobs like this
Is this your posting? Email [email protected] to have it removed.