Founding AI Engineer
Applied AI / ML — the harness around rented models
India, full time remote · Start Sep 2026 · Reports to Founders ·

Atlas, and why this role exists
Atlas is the discovery and intelligence platform for tax transformation — it owns the discovery and readiness phase upstream of implementation, then hands off. Agents read a client’s documents, data and ERP config, and validate findings against sample data at record level rather than workshop assertions. Everything is evidence-chained (record → rule → finding → owner → requirement → test case → sign-off), versioned, and human-signed. Rule packs make it repeatable across mandates.

Rules are deterministic, ML runs on messy evidence, humans hold judgment.
Any competitor can call the same APIs we call. Nobody else has our context — the validated evidence, the expert corrections, the decision graph. You exist to make that context captured, verifiable and trainable from day one, because none of it can be retrofitted later.

What you own
The record
Raw payload archive:



every model request and response stored verbatim from day one, with stable question IDs that survive re-runs.
Model-router layer — "any LLM in, same contract out". Swapping providers must never break a downstream consumer.
The learning log
Every accept, reject, edit and conflict resolution captured in trainable form, with actor and timestamp — labelled training data from the first engagement, not a cleanup project in year two.
The countercheck
Chunk-level attribution and quote-to-source verification. A second model verifies that the quoted source supports the answer; failures die internally and never reach a human.
Confirmed-answer locking with flag-not-overwrite: re-runs diff against confirmed state and raise flags carrying both sources, with the full decision trail.
The eval suite
Task-specific eval sets from real engagements (extraction accuracy, citation validity, conflict-detection recall), with a regression suite wired into every prom

📌 Artificial Intelligence Engineer (India)
🏢 OneSig
📍 India