About JARK Technologies

JARK Technologies builds AI and IoT products. We are a small, senior team, and you will have direct access to our founders and our CTO.

What we are building

We are building a physician-led AI wellness companion. Members talk to an AI companion about their health goals, move through a guided onboarding and a personal journey, and are matched with partner clinics for consultations and follow-up. The platform has a member mobile app, a web portal for partner clinics, and the AI chat and integration services behind them. We are taking it from a validated prototype to a production launch in a HIPAA-regulated setting.

The role

You will own the intelligence and reliability of the products' AI chat: how the companion retrieves clinician-reviewed knowledge, remembers members, stays safe, and streams answers reliably in production.

What you will own

1. Retrieval and knowledge: a RAG pipeline over clinician-reviewed content, covering ingestion, chunking, embeddings, search and reranking, citations, versioning, a review workflow, and a explicit “I don’t know” behaviour.
2. Conversation quality: prompt and persona architecture with version control and review, multi-turn context and memory (what to remember, summarise and forget), and follow-up question behaviour.
3. • Evaluation: a maintained evaluation set of scripted journeys plus adversarial and edge cases, automated scoring with rubrics and calibrated LLM judges, and regression runs that gate every prompt, model or retrieval change.
4. Safety and guardrails: deterministic red-flag detection alongside classifier layers, crisis handling and escalation to clinicians, scope limits (no diagnosis or dosing outside approved scope), prompt-injection and misuse defences, and strict control of which tools the model may call and when.
5. Chat service:



a Python streaming API, durable message storage, idempotent sends, multi-device sync, retries, rate limiting, graceful behavior when a model is unavailable, and per-turn cost controls.
6. Models: benchmarking of hosted and open-weight models for quality, latency and cost, fallbacks, migration plans, and tool-call reliability.
7. Observability: per-turn tracing (prompt version, model, retrieved sources, tokens, latency, safety flags), dashboards, and a loop that turns “not helpful” feedback into new evaluation cases.
8. Clinical collaboration and privacy: turn physicians’ guidance into testable behaviors and a sign-off process, and make sure PHI only reaches vendors covered by signed agreements, with redaction in logs.

What we are looking for

1. 2+ years of software engineering with strong Python: async web services, type hints and automated testing.
2. 2+ years shipping LLM or RAG features to production: embeddings, vector search (for example pgvector or a dedicated store), chunking, reranking and retrieval evaluation.
3. Tool or function calling and streaming responses.
4. Evaluation design: offline evaluation sets, rubric scoring, and knowing when a metric is misleading.
5. Prompt work under version control, treated as engineering and not as tinkering.
6. The failure modes of LLM systems: hallucination, prompt injection and drift.
7. PostgreSQL and API design, and latency and cost optimization.
8. Clear writing for non-engineers,



including physicians.

Nice to have

1. Health or other regulated domains, and HIPAA awareness.
2. Safety classification and guardrail frameworks.
3. Fine-tuning or distillation, and deploying open-weight models.
4. Conversation design experience.
5. LLM tracing and observability tools.

How we work

Weekly sprints, architecture set with the CTO, physician review of clinical behaviour, and changes shipped only when the evaluation set agrees. You will work daily with the full-stack developer and the QA engineer.

What success looks like

1. Measured, not anecdotal, chat quality that improves release over release.
2. Answers grounded in clinician-reviewed sources, with citations.
3. Safety behavior that clinicians trust and that stays stable across model changes.
4. A chat service that never loses a message and degrades gracefully.

How to apply:

Before you submit, please make sure your resume or LinkedIn profile includes links that show your work:

- Your GitHub profile, with the repositories you’d most like us to look at
- Any RAG, evaluation or LLM projects, demos, or technical write-ups you’ve published

You can add the links to your resume, or to the website field or Featured section of your LinkedIn profile. Applications without links may not be reviewed.

Prefer email? Send your resume and a short cover letter to [email protected] with “Senior Python Developer (RAG and AI Chat)” in the subject line.

Equal opportunity at JARK:

JARK is an equal opportunity employer. We hire for skill, craft, and ownership, and we welcome applicants of every gender, background, caste, religion, age, ability, and identity. Career returners and people from under represented communities are especially encouraged to apply.

📌 Python Developer (RAG and AI Chat) (India)
🏢 JARK Technology
📍 India