$240K AI role requires parsing CAS numbers from PDFs

Inside the Funnel

Poka Labs' recruiter screen lasts 15 to 30 minutes and covers four things: experience, salary expectations, availability, and motivation. That is the entire first gate. No whiteboard. No take-home. A five-person Y Combinator startup deployed inside major manufacturers and distributors, including Fortune 100 companies, cannot afford a bloated pipeline — every interview slot burns runway, and every hire either accelerates the product or stalls it. Glassdoor reports the full cycle at roughly two weeks, compressed by design.

What distinguishes this funnel is what it screens for. Poka Labs builds AI agents that ingest messy RFQ emails, parse CAS numbers from PDF attachments, enforce tiered discount logic against live ERP data, and output quotes in minutes — all while keeping a human in the loop. The engineers they hire need to ship that stack end-to-end: React and TypeScript on the front end, AWS infrastructure, LLM orchestration, and the gritty work of normalizing unstructured industrial data.

Public interview accounts stop at the recruiter screen. But the product's architecture implies the assessment: can you build reliable software on top of unreliable inputs? Can you ship fast without breaking the compliance posture that lets you sit inside a Fortune 100 environment? The two-week timeline suggests the team decides quickly once technical signal appears. They're not looking for perfection on a whiteboard. They're looking for evidence you've wrestled the same demons their customers throw at the platform every day.

That evidence is what the next sections map — role by role, skill by skill, screen by screen.

Four Roles, One Core Challenge

Poka Labs' four open positions map directly to the company's core challenge: translating decades of tribal manufacturing knowledge into AI agents that can price, quote, and cross-sell across catalogs exceeding 20,000 SKUs. Each role demands a different blend of software engineering depth, AI fluency, and industrial domain awareness.

Fullstack / AI Engineer (Entry Level)

The posting emphasizes ownership of "large, ambiguous problem areas" and shipping "production AI agents used daily by sales teams, operators, and plant managers." Candidates need full-stack competence in React, TypeScript, and AWS, plus hands-on experience with LLM orchestration. The domain hook is explicit: the agents operate in specialty chemicals, pipe fittings, abrasives, advanced materials, adhesives, coatings, and metals, where technical specifications drive pricing and the best account managers carry decades of unwritten knowledge.

Fullstack / AI Engineer (Senior)

The scope shifts from shipping features to architecting the agent platform itself. The job description cites deployment at "major manufacturers and distributors, including Fortune 100 companies," which means the senior engineer must handle multi-tenant data isolation, audit trails, and the constraints of on-prem or VPC deployments common in chemical conglomerates.

Founding Deployment Strategist

This role sits at the intersection of forward-deployed engineering, solutions architecture, and technical sales. The strategist works with the customer to map the existing quoting process, identify the tribal knowledge holders, and design the agent's decision logic alongside them. The product automates pricing, quoting, and cross-sell for complex manufacturing businesses where catalogs run to 20,000+ SKUs. The strategist must be fluent enough in chemical pricing models (tiered volume discounts, raw-material index pass-throughs, contract-specific overrides) to earn credibility with pricing managers, and technical enough to prototype the agent's logic in code before handing off to the engineering team. The "founding" prefix signals this person will define the deployment playbook, not just execute it.

Product Designer (Contract)

According to Y Combinator's posting, the contract designer role ($4K–$7K monthly, remote-eligible) focuses on the interface layer where sales reps, operators, and plant managers interact with the agents. The challenge is cognitive: rendering probabilistic AI outputs into a UI that a rep trusts during a live customer call. Users include three personas with divergent mental models and risk tolerances. A designer who has only built consumer or generic B2B dashboards will miss the domain nuance: a plant manager needs to see the safety-data-sheet implication of a substitute material; a sales rep needs the one-click "send quote" button that logs the interaction back to the CRM. The contract term suggests a concentrated sprint to establish the design system and the first high-stakes workflows.

The Common Thread

Across all four roles, the non-negotiable is the ability to operate in the gap between clean AI benchmarks and messy industrial reality. The $6 trillion chemical industry "still relies on email and Excel," per the company's own description. Every hire must demonstrate they have built something that survives contact with unstructured PDFs, inconsistent units of measure, and stakeholders who measure success in margin points.

Where AI Meets the Plant Floor

The technical stack implied by the product description is revealing. Ingestion handles Excel, CSV, PDF, scanned technical data sheets, and direct ERP connections without requiring data cleanup — a constraint that pushes the engineering toward robust multimodal parsing and fuzzy matching rather than clean-tabular assumptions. The pricing engine ingests complex spreadsheet logic (freight, tariffs, volume tiers) and can also run an AI-native strategy that prices dynamically off real-time inventory, competitor signals, and historical win rates. Agents route quotes by customer, territory, product category, or deal size, support round-robin load balancing, and flag any quote below a minimum gross-margin percentage for manager approval. All of this runs 24/7 as long-running agents, not batch jobs, with human-in-the-loop review for high-value decisions.

Building and maintaining that architecture calls for engineers who have shipped production LLM systems with tool use, structured output, and evaluation pipelines — and who also grasp why a chemical distributor's "tribal knowledge" about substitution rules and negotiation patterns cannot be modeled as a simple lookup table. On the domain side, the company's own messaging anchors the problem in chemical distribution: customers send vague descriptions, not standard forms; SKUs vary by grade and packaging; pricing rules are layered, exception-ridden, and often undocumented. The agents must learn from historical email context and past quotes, absorbing the same judgment the best reps apply. That means candidates need operational intuition: what a margin threshold means in practice, how approval workflows actually function on a sales floor, and why "no rip-and-replace" is a non-negotiable constraint for a customer running a legacy ERP.

The founding team's background (Harvard School of Engineering, Y Combinator, Parker, Meta, Harvard Business School) signals that the company expects this hybrid fluency at a senior level: people who have built AI systems that touch revenue-critical workflows in regulated, physical-industry settings. The rarity of this profile is the hiring signal. A pure ML researcher who has never integrated with an ERP will struggle to design the agent-to-ERP handoff that respects existing data contracts. A veteran industrial-software engineer who treats LLMs as black boxes will miss the evaluation and guardrail work that keeps pricing agents from hallucinating a discount tier. Poka Labs' four roles are effectively a search for the overlap — engineers and product people who can speak both the language of transformer architectures and the language of gross-margin defense, CAS-number resolution, and approval escalation paths. That overlap is where the product lives, and it is where the screen filters hardest.

Product, Market, and the Hiring Roadmap

Poka Labs sits at the intersection of two hard problems: industrial commerce runs on unstructured chaos, and the margin left on the table is measurable in real dollars. The company's product, an AI commercial operations platform for industrial B2B, automates price setting, quote generation from messy RFQs, and guided selling, all while keeping humans in the loop for approval. Its website claims quotes sent in four minutes, and margin impact tracked at the SKU level. The four open roles map directly to the three product pillars and the infrastructure that binds them.

The price-setting agent ingests cost trends, flags repricing priorities, and sets defensible prices across entire catalogs. The quoting agent parses CAS numbers, grades, quantities, and pricing from emails with PDF attachments, inline images, and vague descriptions, resolving them to exact SKUs. The guided-selling agent surfaces TCO analysis, margin defense talking points, and cross-sell recommendations. Each agent connects to the customer's existing ERP, CRM, and spreadsheets (Salesforce, NetSuite, SAP, Excel) without rip-and-replace. The platform also ingests scanned TDS and MSDS sheets, learns from historical quotes, and enforces tiered discounts, customer-specific rates, and approval workflows. Security is enterprise-grade: AES-256 at rest, TLS 1.2+ in transit, regular third-party penetration testing.

Market forecasts vary widely. One firm projects the industrial AI market at $7.1 billion in 2025, growing to $150 billion by 2033 at a 46.5% CAGR; another puts it at $12.5 billion reaching $40.5 billion by 2034 at 14.5% CAGR; a third sees $280 billion by 2035 at 46% CAGR. The spread reflects different definitions of "industrial AI," but the trajectory is consistent: industrial buyers are moving from pilot to production, and they need platforms that handle their data as it exists — messy, fragmented, and rule-heavy.

Poka Labs' go-to-market reflects that reality. The company pairs software with pricing consulting engagements: experts work alongside the platform to find hidden margin, reprice catalogs, and build commercial operations that scale. The stated goal is to train the customer's team, capture tribal knowledge in the platform, and hand over the keys. Deployments target days, not months. An aimyflow analysis notes the product is positioned as a layer atop existing commercial systems rather than a replacement, and that success depends heavily on how completely pricing logic, approval thresholds, and substitution rules are captured. The same analysis recommends starting with a narrow workflow, such as one quoting process, product line, or sales segment, to test accuracy and adoption before expanding.

The hiring plan mirrors this roadmap. The full-stack engineer builds the integration layer that connects to legacy ERPs and spreadsheets without breaking them. The product designer shapes the human-in-the-loop review surfaces where margin-sensitive decisions get approved or overridden. The founding GTM hire translates the margin-impact story (Product A Poka Labs found +4.2%, Product B -1.8%, Poka Labs' figures put $142K TCO savings identified) into a repeatable sales motion for industrial distributors and manufacturers.

A broader agent ecosystem is already visible. An aimyflow workflow sketch combines Poka Labs with procurement, account research, and post-quote follow-up agents: one extracts RFQ details, another checks account history and margin thresholds, another prepares seller talking points, another tracks quote status and next actions. That vision, coordinated agents across the commercial lifecycle, is what the current roles are building toward.

The industrial AI wave is priced in the market forecasts, embedded in the RFQs landing in distributor inboxes today, and measured in the margin points Poka Labs' demo data already shows. The four roles exist because the product has moved past proof-of-concept into the grind of production hardening, integration depth, and go-to-market execution. Whoever fills them will ship the agents that turn messy industrial commerce into policy-driven, margin-aware operations — one quote at a time.

What the Interviewers Actually Score

Poka Labs runs a process that begins with a recruiter screen covering background, career goals, role alignment, and salary expectations. Glassdoor reports this call is a fit and alignment check. Come with a clear, consistent narrative about why Poka Labs, why now, and what you want next. Vague answers on compensation expectations or career direction create friction early. The company has no H-1B petition history, so visa sponsorship is unlikely — factor that into your planning if it applies.

Technical rounds assess implementation details and strategic mindset. For the Fullstack / AI Engineer role, be ready to discuss how you would architect systems that handle messy real-world inputs, since the product promises "commercial operations that run themselves" with ambient agents that price, quote, and guide selling 24/7. For the Founding Deployment Strategist and Founding GTM roles, the emphasis shifts to stakeholder management. Prepare concrete examples of driving outcomes through cross-functional work.

The process is remote via Zoom. Test your setup, lighting, and audio beforehand. Demonstrate you understand structured evaluation, tradeoffs, and decision-making. Ask questions about the process itself; it shows you think in systems.

HR discussions cover compensation structures, career growth, and organizational alignment. Do not wing these. The posted base range spans $120K to $240K depending on role and experience, and equity can be meaningful at a successful startup. Know your number, know the market, and be ready to discuss how the role fits your trajectory.

Finally, research the company. Read the mission statement. Understand the product — "ambient agents that build prices, automate quotes, and guide selling to defend your margin." Prepare questions that reflect genuine interest in the problem space. Candidates who align their preparation to communication, leadership, process thinking, strategic framing, and honest self-assessment are the ones who convert.

What This Signals for Frontier-Tech Hiring

Poka Labs' search for engineers who can bridge AI and industrial operations mirrors a shift that has already remade the earliest stage of the venture market. Y Combinator's Summer 2025 batch was 88 percent AI-native (141 of 160 startups), and the Fall 2025 cohort doubled down on agent orchestration, developer productivity, and enterprise automation. The accelerator's data shows founding teams averaging 6.5 years of professional experience; 48 startups have alumni from Google, Meta, or Amazon. First-time founders with under a year of experience barely register. The message is clear: investors are betting on operators who have already shipped, not on raw potential.

That bias trickles down to hiring. Garry Tan, YC's CEO, said the current batch is growing 10 percent week over week in aggregate, a pace he called unprecedented in early-stage venture. The lever is AI-assisted coding. "Vibe coding," Tan's term for letting models generate entire features, lets teams of fewer than ten people reach $10 million in revenue. Capital goes further. Headcount stays flat. Every hire must carry disproportionate weight. Poka Labs' four roles, each demanding both model fluency and floor-level domain knowledge, fit that template exactly.

The screening side is automating in parallel. Alex, an AI recruiter that raised a $17 million Series A led by Peak XV Partners, now conducts thousands of voice interviews a day. Its founder, Aaron Wang, told TechCrunch the goal is to build a professional profile deeper than LinkedIn from a ten-minute conversation. Mercor, another AI recruiter turned data-labeling platform, is chasing a $10 billion valuation. Job seekers should expect more initial screens, not more open roles. The funnel widens at the top; the gate tightens.

At the other end of the spectrum, the talent war has gone nuclear. Meta offered $100 million signing bonuses to OpenAI staff, acquired a 49 percent stake in Scale AI for $14.3 billion to absorb Alexandr Wang and his engineers, and pursued Safe Superintelligence's Daniel Gross after a $32 billion valuation blocked a full buyout. Google re-acquired Character.AI's founders; Microsoft paid $650 million for Inflection's talent. OpenAI spent $6.5 billion on Jony Ive's device startup. These are not hiring budgets — they are strategic M&A disguised as payroll.

Poka Labs cannot compete on that scale. It doesn't need to. The frontier-tech labor market is bifurcating: a thin layer of celebrity researchers commanding nine-figure packages, and a broadening base of hybrid practitioners who can deploy models inside real workflows. The latter group, exactly what Poka's ads describe, is where the next wave of enterprise value gets built. YC's Fall 2025 themes confirm it: multi-agent orchestration infrastructure, developer productivity platforms with measurable impact, generative video with latency and cost advantages. All require the same blend Poka screens for: model competence plus production rigor.

An investor screening rubric circulating in seed circles weights team at 20 points, product and tech moat at 25, traction at 20, unit economics at 15, market size and timing at 10. Legal and compliance are penalty-only. Notice what's missing: pedigree, headcount plans, hiring brand. The market rewards evidence of execution. Poka's process (technical screen, domain exercise, team interview) is a microcosm of that rubric.

Anthropic CEO Dario Amodei predicted AI could eliminate half of entry-level white-collar jobs in one to five years, pushing unemployment toward 20 percent. The same forces creating that risk are creating the roles Poka lists. Someone has to build the systems that replace the routine work, and someone has to maintain them when the model hallucinates a spec. That is the hiring signal. It is not about headcount growth. It is about leverage per head.

The recruiter screen still lasts 15 to 30 minutes. It still covers four things. But the candidates who clear it now know exactly what the next room looks like: a pricing manager watches an agent parse a CAS number from a PDF, and the engineer beside them has already built the guardrail that keeps the margin threshold intact.

Working in AI? Zero G Talent tracks the openings: see every open Databricks role, browse AI jobs, openings at Anthropic and Harvey AI, and the people building the field.