Zep AI (YC W24) Is Hiring a Head of Forward Deployed Engineering

Marketing Manager

San Francisco, CA, US / Remote (US)·Full-time·Marketing·3+ years

$120K - $180K0.15% - 0.35% equity

AI Agents

Zep manages, governs, and serves supplier representation astatine endeavor scale. Enterprises build connected Zep to tally reliable, personalized agents crossed the business: millions of Context Graphs, served successful nether 200ms, wrong their ain VPCs and unreality deployments. Customers see Samsung, Zscaler, Twin Health, HoneyBook, and NASDAQ 100 and Fortune 500 exertion companies. We besides build Graphiti, our open-source discourse chart model (30K+ GitHub stars).

You will ain really engineering leaders and developers find Zep, understand it, and determine to build connected it: hunt and reply engines, the blog and website, email, and social. The founders group the contented strategy and agents do the accumulation work. You tally publication and the campaigns astir it.

We're hiring a marketer who runs endeavor B2B request procreation unsocial and uses agents to do the activity of a team. There is nary trading squad to inherit and nary to manage.

We will measurement you connected signups and connected the pipeline your campaigns produce.

You'll study to our founder, Daniel (2x founder, engineer, erstwhile caput of ML astatine SparkPost), and subordinate a squad pinch pedigree astatine Scale AI, Dropbox, ActiveCampaign, DroneDeploy, and McKinsey.

How we work

We're a small, distributed squad that useful intimately together. We brace connected difficult problems, reappraisal each other's designs, and dainty learning arsenic portion of the occupation alternatively than thing that happens aft hours. We inquire a batch of questions: of customers, of teammates, of our ain assumptions. When we find pain, we spell hole it.

We expect the aforesaid back: inquire questions early, push backmost erstwhile you disagree, and attraction astir the group connected the different extremity of the API.

What you'll do

• Own hunt and reply motor discoverability. Buyers inquire Claude and ChatGPT earlier they inquire Google, and their coding agents inquire connected their behalf. Structured data, llms.txt, tract and docs architecture, and the measurement that tells america whether immoderate of it works.
• Run the blog and website publication pipeline. Content strategy and overmuch of the long-form penning travel from the founders. You tally the almanac and accommodate each draught to its transmission earlier it publishes connected schedule.
• Run email: lifecycle from signup to income conversation, and outbound campaigns into endeavor accounts alongside sales.
• Run societal successful the places engineers and their leaders are, utilizing agents to move accelerated and your ain sensation to determine what ships.
• Build and run the agents that nutrient content, tally campaigns, show competitors, and study results. You ain them aft they ship.
• Own the chimney measurement from first touch to signup, pinch our engineers connected the instrumentation. Choose the stack and support the costs of your ain automation successful view.

Your first 90 days

You commencement by publishing. By time 30 the blog and societal cadence is moving connected an supplier pipeline you built. By time 60 the chimney is measured from first touch to signup and you person a hunt and reply motor baseline to move. By time 90 you ain the request procreation number and person a roadmap you tin defend.

What we're looking for

1. You person tally B2B request procreation for a method merchandise and tin show the pipeline it produced.
2. You edit good and constitute erstwhile needed. Show america thing you published that engineers publication to the end.
3. You build pinch agents arsenic a matter of routine. Claude Code, Cursor, aliases balanced to nutrient contented and tally campaigns, arsenic portion of really you activity alternatively than thing you person tried.
4. You are comfortable successful code: editing the tract aliases wiring an integration. You do not request to person shipped applications.
5. You want to tally a usability alone, and you dainty a squad of 1 arsenic an advantage.

Useful, not required: anterior trading astatine a dev tools, infrastructure, aliases AI company. SEO and AEO acquisition grounded successful tests alternatively than theory. Paid acquisition. Familiarity pinch supplier frameworks, retrieval, aliases chart databases.

This domiciled is astir apt NOT a fresh if:

• You tally campaigns done agencies and contractors and person not done the accumulation activity yourself.
• You person not built pinch coding agents and do not intend to start.
• You want to prosecute a squad earlier you person built the systems.
• You request an established playbook aliases sign-off earlier you move.

Compensation and benefits

Platinum medical, dental, and imagination insurance. 401K pinch employer matching. Unlimited PTO. Flexible in-office civilization successful San Francisco, pinch distant options and periodic recreation for squad members extracurricular the Bay Area.

How we hire

We respect your clip and support our question and reply process tight and focussed.

Screening Call (w/ Daniel, our Founder) → Team Calls (2-3 hours back-to-back, includes a portfolio walkthrough) → Decision Call (Daniel, again)

How to apply

Pick 1 run aliases contented programme you ran extremity to end. What pipeline did it produce, and what would you trim adjacent time? (150 words aliases fewer.)

Member of Technical Staff: Research

San Francisco, CA, US / Remote (US)·Full-time·Engineering·Machine learning·6+ years

$180K - $250K1.00% - 1.50% equity

Amazon Web Services (AWS)PythonTorch/PyTorchMachine LearningLLMsAI Agents

Zep manages, governs, and serves supplier representation astatine endeavor scale. Enterprises build connected Zep to tally reliable, personalized agents crossed the business: millions of Context Graphs, served successful nether 200ms, wrong their ain VPCs and unreality deployments. Customers see Samsung, Zscaler, Twin Health, HoneyBook, and NASDAQ 100 and Fortune 500 exertion companies. We besides build Graphiti, our open-source discourse chart model (30K+ GitHub stars).

What our customers' agents tin logic astir depends connected the representation we retrieve and the representation we write. You ain that loop. You build agents that amended retrieval. You finetune the models that extract memory, and the models that powerfulness those agents. You tally the experiments and vessel the consequence arsenic accumulation code.

We're hiring an technologist who builds agents and trains the models they tally on.

We will measurement you connected whether retrieval and representation value move, and connected what reaches production.

You'll study to our founder, Daniel (2x founder, engineer, erstwhile caput of ML astatine SparkPost), and subordinate a squad pinch pedigree astatine Scale AI, Dropbox, ActiveCampaign, DroneDeploy, and McKinsey.

How we work

We're a small, distributed squad that useful intimately together. We brace connected difficult problems, reappraisal each other's designs, and dainty learning arsenic portion of the occupation alternatively than thing that happens aft hours. We inquire a batch of questions: of customers, of teammates, of our ain assumptions. When we find pain, we spell hole it.

We expect the aforesaid back: inquire questions early, push backmost erstwhile you disagree, and attraction astir the group connected the different extremity of the API.

What you'll do

• Build agents that amended retrieval: query understanding, ranking, and what to propulsion into context.
• Finetune the models that extract, update, and consolidate representation connected Zep's domain.
• Finetune the models that powerfulness those agents, truthful the retrieval loop and the supplier get amended together.
• Own the activity from dataset creation done research design, evaluation, training, and a alteration that ships to production. Our engineers activity the serving way alongside you.
• Build the eval harnesses that drawback regressions successful retrieval, representation quality, and supplier task completion earlier a merchandise ships.
• Write up what you find, including the results that killed an idea, truthful the remainder of engineering tin build connected it.

Your first 90 days

In your first week you prime up an unfastened retrieval aliases memory-extraction problem and framework it arsenic an experiment. By time 30 an supplier aliases a finetune you built is successful the merchandise loop, aliases you person ruled the attack retired and written up why. By time 90 you are 3 cycles successful and the eval harness you built runs connected each release.

What we're looking for

• You person shipped accumulation ranking, retrieval, aliases query understanding.
• You've built classical ML into a existent merchandise — logistic regression, SVMs, GBTs, single-layer perceptrons.
• You've shipped finetuned models to production. You cognize transformer architectures and training workflows, and you activity successful PyTorch.
• You person shipped a non-trivial agentic strategy to production. Not a prototype, not a bladed wrapper complete a chat-completion API.
• You person a investigation methodology: dataset creation and curation, research design, and evaluation. You tin framework an unfastened problem and creation an research that answers it.
• You person built information for retrieval, generation, aliases supplier tasks: golden sets, offline metrics, online tests.
• You constitute beardown Python, and you person capable accumulation AWS to ship, monitor, and iterate connected what you train.
• You person a Master's successful Computer Science aliases balanced experience.

This domiciled is astir apt NOT a fresh if:

• Your shipped activity is papers, prototypes, and demos.
• Your supplier activity is simply a wrapper complete a chat-completion API, pinch nary eval and nary exemplary you trained.
• You haven't finetuned a exemplary that reached production.
• You request a investigation schedule handed to you earlier you move.

Interview process

Head of Forward Deployed Engineering

San Francisco, CA, US / Remote (US)·Full-time·Engineering·Full stack·11+ years

$220K - $270K1.20% - 1.75% equity

Amazon Web Services (AWS)GoKubernetesPythonTypeScriptTerraformAI Agents

Zep manages, governs, and serves supplier representation astatine endeavor scale. Enterprises build connected Zep to tally reliable, personalized agents crossed the business: millions of Context Graphs, served successful nether 200ms, wrong their ain VPCs and unreality deployments. Customers see Zscaler, Samsung, Twin Health, HoneyBook, and NASDAQ 100 and Fortune 500 exertion companies. We besides build Graphiti, our open-source discourse chart model (30K+ GitHub stars). Getting Zep into accumulation successful customer environments is its ain engineering problem, and you ain it.

You will lead our deployments and build the squad that runs them. The first twelvemonth is hands-on. You will beryllium successful architecture reviews and successful our codebase, and connected tract erstwhile a deployment needs you. Our engineers activity deployments alongside you, truthful you are ne'er the only method personification connected the call.

We're hiring an technologist and leader who has done this activity and tin now group the modular for really we do it.

This is statement a halfway merchandise engineering role. You should person important customer-facing acquisition successful an FDE, income engineering, aliases section engineering role.

We will measurement you connected clip from signature to production, and connected civilization engineering per deployment.

You’ll study to our founder, Daniel (2x founder, engineer, erstwhile caput of ML astatine SparkPost), and subordinate a squad pinch pedigree astatine Scale AI, Dropbox, ActiveCampaign, DroneDeploy, and McKinsey.

How we work

We're a small, distributed squad that useful intimately together. We brace connected difficult problems, reappraisal each other's designs, and dainty learning arsenic portion of the occupation alternatively than thing that happens aft hours. We inquire a batch of questions: of customers, of teammates, of our ain assumptions. When we find pain, we spell hole it.

We expect the aforesaid back: inquire questions early, push backmost erstwhile you disagree, and attraction astir the group connected the different extremity of the API.

What you'll do

• Own the method result for our largest accounts, from the first architecture speech done 90 days successful production.
• Write code. Integration work, reference architectures, deployment tooling, and fixes successful the merchandise itself.
• Deploy Zep wrong customer infrastructure crossed Managed, BYOK, and BYOC, and activity straight pinch the level and information teams who person to o.k. it.
• Define the engagement model. Decide which accounts get embedded support, and group really that activity is scoped and priced.
• Hire the team. You specify the barroom and lead the engineers you bring in.
• Reduce the civilization activity each deployment takes. What you build doubly becomes a reference architecture aliases a merchandise requirement.
• Take what you study successful accumulation backmost into the roadmap. You will beryllium successful readying pinch the engineers who tin enactment connected what you bring.

Your first 90 days

You commencement wrong a unrecorded deployment: shipping code, gathering the customers we already serve, and learning really Zep runs successful their environments. By time 60 you ain each progressive deployment. By time 90 you person written the first type of the engagement model.

\What we're looking for

• 10+ years crossed package engineering, guardant deployed engineering, solutions architecture, aliases income engineering, pinch respective of those years arsenic a hands-on engineer.
• You person taken endeavor deployments from first telephone to accumulation nether a information reappraisal and a fixed deadline.
• 3+ years starring customer-facing method people, aliases a clear lawsuit for why you are fresh to.
• Strong Python, positive Go aliases TypeScript. You will publication our codebase and nonstop patches.
• You person shipped a non-trivial agentic strategy to production. Not a prototype, not a bladed wrapper complete a chat-completion API.
• Comfortable successful customer infrastructure: AWS, Kubernetes, Terraform, and the personality and encryption activity a BYOC instal requires.
• You raise risks early, including erstwhile it is inconvenient for the personification you are talking to.

This domiciled is astir apt NOT a fresh if:

• You person tally a master services P&L and deliberation of this arsenic a transportation org.
• Your customer-facing activity ends pinch the method triumph and a handoff astatine signature.
• You haven't written accumulation codification successful respective years and don't intend to commencement again.
• You request an established playbook aliases sign-off earlier you move.

Interview process