ATACANA GROUP Inc.
Data Engineer — AI Platform, Extraction & Distributed Systems
Get notified of more jobs like this · No spam, ever
How your CV stacks up
Upload your CV to see how well it fits this job role
?%
About Atacana
Atacana Group, Inc. is a global competitive strategy and intelligence consulting organization focused on the healthcare industry. We build and operate our own AI-powered intelligence platform, Avizum, it is currently used daily by professionals at large pharmaceutical companies. This role sits on that product team.
Why join our fast-growing team?
- Build real AI, not demos. AI systems running in production against proprietary data — with real users, real SLAs and real feedback loops.
- Work that matters. The decisions our platform informs shape how treatments are developed, positioned and brought to patients who need them.
- Results driven. We measure output, not when you work. Flexible hours let us schedule our days so that we do our best work without missing out on life’s important moments.
- Freedom of location. 100% remote work. We are a globally distributed team, so we can work wherever we are happiest.
- Diverse and global team. Our team members are located across 4 continents and 15+ countries.
- Growth mindset. Atacana is a learning organization. We stay close to the latest research and tools, share what we learn, and expect everyone to keep leveling up.
The role in one paragraph
We are building the next generation of our platform. This role is accountable for the data it runs on: acquiring content — increasingly including audio and video — from a large and growing set of external sources, extracting structure from it reliably, and delivering it into downstream services with the quality, freshness and provenance our clients expect. It is primarily a hands-on engineering role with real architectural ownership, working alongside AI/ML, backend, data and infrastructure engineers.
Stack. Python-centred and cloud-native: asynchronous services, containers and orchestration, relational and specialised data stores, and a mix of commercial and self-hosted language models.
What you’ll own
You will be building the system that turns the open, messy, uncooperative web data into a corpus that pharma executives, healthcare professionals, patients and researchers can trust and cite.
- Own the crawler and extraction fleet. Design, scale and operate a large portfolio of crawlers and extractors across many heterogeneous external sources — scheduling, politeness and rate limits, incremental crawling, deduplication, and full provenance from raw artefact to published record. You will shape this system from architecture down to the code.
- Keep extraction alive as the web moves. Sources restructure and paginate differently. Build the monitoring, alerting and self-healing that catches silent breakage. Drive the time from breakage to fix toward zero. It is an adversarial, always-changing problem, which is what makes it interesting.
- Turn any format into structured data. HTML, PDFs, tables and images, and increasingly audio and video. Combine deterministic parsing with model-based extraction where it earns its place — structured outputs, cost control and human review where accuracy demands it — and measure the accuracy of both.
- Own data quality as SLOs. Coverage, freshness, completeness and duplication as measured, alerted thresholds — not as anecdotes. Build the contracts and validation that stop bad data reaching downstream consumers, and set the standard the rest of the company trusts by default.
- Ship changes on evidence, not vibes. Stand up evaluation harnesses, golden datasets, regression suites and tracing, so that every model, prompt and parser change is a measured improvement rather than a hopeful one.
- Own the data platform end to end. Boundaries, contracts, data ownership, failure modes and migration paths; pipelines that are event-driven, idempotent, resumable, observable and graceful under partial failure or provider outage. Containers, CI/CD, infrastructure as code, cost and latency budgets, tenant isolation and data security are part of the job, not someone else’s. Write the design docs and defend the trade-offs.
- Make spoken content first-class. Acquire and process audio and video at scale: transcription, speaker attribution, alignment of transcripts to source media, and extraction of the substance from long recordings. Conference talks, panels, and briefings that are effectively invisible today become part of the corpus because you made them so.
Reasons to use Rodeo
Start with a chat, not a search bar
Grad scheme, placement, apprenticeship? Not sure what you want yet — that's fine. Your agent talks it through with you and turns "I have no idea" into a shortlist.
It searches the market for you
Every day your agent scans the market matching roles against what actually matters to you, not just keywords on a CV.
Only hits
No noise. No "maybe this fits." Just roles with a clear explanation of why they're right — and where to focus when applying.
What we expect at this level
Beyond delivery, we expect you to raise the bar of the team: review code and designs, mentor engineers, and translate ambiguous business intent into a technical plan. Your influence should outlast any single system you build here.
What we’re looking for
If you meet most of the essentials below and are missing one or two, we would still like to hear from you.
Data extraction & crawling at scale
Essential
- Has built and operated a non-trivial crawler fleet — hundreds of sources — over a sustained period, not a one-off scraping project.
- Practical experience of the hard parts: politeness and rate limiting, session and cookie handling, proxy and IP management, pagination and incremental crawling, change detection, retry semantics, and rendering JavaScript-heavy pages reliably.
- Treats parser drift as a first-class engineering problem: schema and selector versioning, contract tests against fixtures, canary sources, and alerting on volume, freshness and null-rate anomalies rather than on exceptions alone.
- Document and content extraction: HTML-to-text normalisation, PDF parsing, tables, boilerplate removal, language detection, near-duplicate detection and content canonicalisation across syndicated sources.
- Backfill and reprocessing strategies for when extraction logic changes retroactively, including lineage that survives a reprocess.
- Working awareness of robots.txt, terms of use, licensed-content restrictions and the difference between what is technically possible and what we are permitted to do. In our industry, where data came from matters as much as what it says.
Also valuable
- Audio and video processing at scale: speech-to-text pipelines (self-hosted or API-based), speaker diarisation, chunking and timestamping of long recordings, handling accents and poor audio, media transcoding and storage cost management — with a clear view of transcription accuracy as a measured quantity rather than an assumption.
- OCR for scanned or image-only sources.
Get help with your application
Your very own career expert that helps elevate your application to the next level.
Software engineering & architecture
Essential
- 7+ years building and operating production software, including meaningful time owning architecture rather than only implementing it.
- Expert-level Python: async/await, typing, testing discipline, packaging, profiling. Strong API design (FastAPI or equivalent) and clean service boundaries.
- Demonstrable experience decomposing a system: modular monolith vs. microservices, contract and schema evolution, backwards-compatible migrations, and decommissioning legacy components without downtime.
- Comfortable writing architecture decision records and defending trade-offs in front of both engineers and non-technical stakeholders.
Distributed systems
Essential
- Solid fundamentals: idempotency, at-least-once vs. exactly-once delivery, retries and backoff, backpressure, poison messages, ordering, eventual consistency, distributed locks, partial failure.
- Hands-on with message brokers and task or workflow orchestration frameworks.
- Containers and orchestration in production: autoscaling, resource limits, health and readiness, rollout strategies, debugging a live incident.
- A relational database at scale — indexing, query plans, partitioning, connection pooling — and practical experience with multi-tenant data isolation.
- Observability as a habit: structured logging, metrics, distributed tracing (OpenTelemetry), SLOs and on-call ownership.
Also valuable
- Experience with a vector database, and ideally a graph database.
Applied AI
Essential
- Real, hands-on delivery of LLM-based systems to production — not prototypes. Retrieval-augmented generation, tool calling, structured output, prompt and context engineering.
- Understands why LLM systems fail at scale: context limits, error accumulation across multi-step workflows, cost and latency blow-ups, and injection risks when processing untrusted external content.
- Has built or owned an evaluation strategy: offline eval sets, human review loops, regression gates in CI, and online quality signals.
- Judgement about where LLMs belong in a pipeline: knows that a rule or a trained classifier is sometimes the right answer, and can justify the cost of a model call per document at production volume.
Also valuable
- Provider-agnostic thinking: has worked with more than one model provider and can reason about routing, fallback, caching, token economics and self-hosted vs. API trade-offs.
Ways of working
- Investigative mindset — able to isolate defects across service, data and model boundaries.
- Excellent written communication; able to prioritise and coordinate work from high-level goals with little supervision, across time zones.
- Fluent in written and spoken English.
“It took my CV and asked me questions relevant to understanding what kind of jobs to suggest for me. Suggestions were almost perfect. Jobs were exactly what I’ve been looking for.”
Jessica, London
Skills
Location