About Archimedis Digital

Archimedis Digital is a global professional IT services company that builds digital products and offers services exclusively to Life Sciences companies. We partner with clients to formulate and deliver growth across the digital life cycle. Leveraging our team's extensive Life

Sciences knowledge and IT expertise, we tackle and solve complex cross-functional business challenges. Our offerings are crafted to make digital transformation a breeze.

About the Role

We're building AI products for the life sciences industry (highly regulated) based on robust AI architecture frameworks & technologies such as knowledge graphs & SLMs. This role owns making these AI systems provably production-ready: reproducible, evaluated, explainable,

and defensible in front of a customer's own auditors. You will be the person who can say,

with evidence, exactly which model version produced a given answer, why it's grounded in the knowledge graph rather than guessed, and that it never learned from a customer's regulated data.

You'll work hands-on with the ML engineering work already underway (Fine-tuning of SLM,

GraphRAG retrieval over a Neo4j knowledge graph) and build the operational discipline

(versioning, evaluation, monitoring, audit trails) that turns a working prototype into something a regulated customer can trust in production.

What you'll own

- Model/prompt/knowledge-graph versioning, so every answer the system ever gave is reproducible against the exact model, prompt, and graph snapshot that produced it
- A golden-set evaluation harness built from the product's competency-question set, run before any model or graph update ships
- Production monitoring for hallucination/grounding,



flagging answers that don't trace back to a retrieved graph fact
- Defining and implementing the system's "I don't have enough context" fallback behaviour, so the product never guesses confidently when it shouldn't
- Drift monitoring as the knowledge graph grows, canary/eval-gated promotion for new model releases, and a model-card/training-run manifest for every release
- The technical wall between production customer logs and the training-data corpus, the actual evidence behind "we never train on your data".
- Supporting the compliance lead on ISO/IEC 42001 control design and DPA language for

AI/training-data use
- Extending audit-trail immutability to cover AI-generated outputs (model version,

inference inputs) alongside human actions
- Prompt-injection and adversarial-input defences and the production workflow for subject-matter-expert review/override of AI outputs
- Governance of the fine-tuning pipeline: reproducible runs, approved-corpus-only inputs, sign-off before any release
- Qualification testing of customer-supplied LLMs as well as our SLM

Who you'll work with

• Daily collaboration with Product Engineering on explainability UI and the override/escalation workflow

• Close coordination with Testing & Validation on the eval harness and golden-set definition (they'll often supply or sign off on the ground truth)





• Periodic coordination with Infrastructure on where monitoring/telemetry

• Direct interaction with AI Governance Committee on what the technical controls need to prove for ISO 42001 and customer DPAs.

What we're looking for

• 4–7 years in software/ML engineering, with at least 1–2 years hands-on with LLMbased systems (fine-tuning, RAG, eval harnesses, or production LLM monitoring)

- Comfortable reading and reasoning about model behavior without needing to have built a foundation model from scratch. Practical MLOps instincts matter more than research depth

• Working familiarity with at least one eval/observability tool in the LLM space (e.g.,

Langfuse, DeepEval, Ragas) or a clear ability to pick one up fast

• Some exposure to regulated or audit-heavy environments (healthcare, finance, life sciences, or similar) is a solid plus, but we'd rather hire someone sharp who can learn

GxP/CSA specifics than someone who knows GxP but can't do the engineering

- Industry certifications in the relevant technology a plus

Soft skills (that matter more than usual here)

• Translates between worlds: can explain to a compliance lead why a model can't be

"proven correct" the way code can, and can explain to an engineer why an auditor's question about model provenance isn't box-ticking

• Comfortable saying "this isn't grounded enough to ship", even under schedule pressure, the role exists specifically to hold that line

• Asks for help across team lines: this role only works by pulling in Testing & Validation,

Infra, and Compliance regularly To Apply:

Please send your resume and a cover letter outlining your relevant experience [email protected]

📌 AI Systems & Compliance Engineer (Tamil Nadu)
🏢 Archimedis Digital
📍 Tamil Nadu