Founding Research Engineer

Specific (YC F25, $4M raised) helps AI teams get high-quality human data across voice, code, language, research, and professional domains. It sells human-generated training data directly to frontier labs: voice recordings, audio conversations, enterprise workflow data, customer service calls, sales calls, and healthcare conversations.

The bet is that the public corpus of internet text is used up, and hiring random contractors does not produce the quality needed for models that can work as coworkers. The next data is decades of enterprise data sets current models have never seen, plus multilingual voice from regions where models are weak, including Arabic and Indian regional languages.

Specific runs a marketplace. On the supply side, partnerships with call centers and a network of data providers across 20+ countries who use Specific's apps and consent to their voice being recorded in exchange for payment. On the demand side, frontier labs set tight data requirements and tight deadlines. Specific is the layer that delivers that data reliably and at quality. Seed stage, about 1 to 10 people, founded in 2025, in San Francisco.

Specific is hiring its first engineer in San Francisco, a founding research engineer, to build the pipeline of coding reinforcement-learning evals and tasks. They are already mid-flight on pilots and contracts with top frontier labs and want engineering in the office before that work outgrows the team. You would be about the fifth person in the room.

Day one you work with customers and on RealSuite: running evals of frontier-lab models on the benchmark and on representative tasks. By about three months you are building RL tasks directly with the labs. By about six months the hope is that you are leading a team of engineers. You own quality control and delivery to spec.

The profile is research and engineering together. You have to understand what an eval is, what a harness is, and what an RL task is, and you have to be able to build them. Already understanding RL, coding RL, evals, harnesses, benchmarks, and post-training is non-negotiable. They will not teach that from scratch. The other non-negotiable is young, high-energy, and hungry. The two traits that worked on their India team were prior experience at a data company and wanting to do and learn as much as possible. Ideal backgrounds include Handshake, Prime Intellect, Mercor, Trajectory, Bespoke, and AfterQuery. Without that, a personal benchmark, a side project, or a post-training experiment counts. Research at a strong lab or school is a bonus. Pure research with no build ability, or a pure engineer with no eval intuition, is a miss. So is showing up for quick cash, low integrity, or low agency.

In office in San Francisco. Visa sponsorship is limited. Equity is about 0.25% to 0.5%. Free lunch and health insurance. Posted work intensity is 9-9-6. Experience they have in mind is about 1 to 5 years. The process is a call with the founder, a culture-fit call with the co-founder, a take-home RL and evals build, then a one-day on-site in San Francisco.