Senior Software Engineer, AI
Practice by Numbers (PBN) | Gurugram, India Scope: AI / Conversational Products & Backend Services
About Practice by Numbers
Practice by Numbers is a dental practice management SaaS platform serving over 1,500 practices across North America — practice management software, VOIP, payment processing, and analytics. We're now expanding into AI-powered automation, building conversational AI products that change how practices interact with their patients.
About the Role
We’re looking for a Senior Software Engineer with strong, hands-on experience in both backend product engineering and modern AI systems. The ideal candidate can design and build reliable, production-grade services while also developing, integrating, and deploying AI-powered capabilities. As this is a single opening on a small team, we need someone who can take end-to-end ownership across both areas and contribute independently throughout the product development lifecycle.
You'll work on our AI Receptionist — a multi-channel conversational AI (voice, SMS, web chat) for dental practices — and the backend services, APIs, and integrations behind it. Hands-on IC role: you own features end to end, including their production behaviour. Real patient-facing traffic under HIPAA constraints, where correctness and latency both matter.
Reports to: Lead Engineer — AI Location: Gurugram, India — this role is open in Gurugram only Work Mode: In-office Working Hours: Primarily IST (10 AM – 5 PM), with some evening overlap with US teams (until 9–11 PM IST) as needed
What You'll Do
Backend
- Build and maintain backend services and RESTful APIs in Python (FastAPI / Django)
- Design schemas and write efficient PostgreSQL queries; use Redis for caching and session state
- Work with async and event-driven patterns — queues, webhooks, WebSockets, background workers
- Own the operational side: logging, metrics, alerting, debugging production issues
- Write unit and integration tests for the business logic you ship
AI & LLM
- Own the conversation layer for real-time voice — turn-taking, barge-in, and recovery when the agent talks over a caller or stalls mid-turn
- Debug non-deterministic failures in production: fabricated confirmations, invented practice policies, dropped context across turns
- Own patient identity resolution over the phone — misheard and misspelled names, shared phone numbers, multi-algorithm phonetic matching
- Version prompts and tool schemas like code: canary changes, measure impact, roll back regressions
- Own the eval and regression suite — golden conversation sets, replay of production calls, CI gating on prompt and model changes
- Design guardrails for patient-facing interactions: no medical advice, no unverified data disclosure, tool-call authorization, PHI excluded from logs and traces, escalation to a human when the agent should stop
- Own the model and voice vendor abstraction so a provider change is configuration, not a rewrite
- Keep voice, SMS, and web chat consistent — shared state and knowledge base, channel- appropriate behaviour
Integrations & Data
- Integrate with practice management systems (Dentrix, Open Dental, Eaglesoft) and internal PBN APIs
- Implement secure auth flows, including OTP-based patient verification
- Define SLOs for conversation success and turn latency; own incident response and postmortems for this surface
Collaboration
- Work with Product Management to turn requirements into working software, surfacing edge cases early
- Participate in sprint planning, standups, code reviews, and product reviews
- Document what you build; collaborate across time zones with US-based stakeholders
Required Qualifications
Experience
- 4+ years of professional software development experience
- Hands-on experience building backend services and APIs that ran in production
- Practical experience with LLM-based applications (GPT-4/4o, Claude, or similar) — prompt design, tool calling, handling model output in real systems. Substantial personal or open-source work counts; tutorial-level does not.
- Experience debugging and improving a system after it shipped
Technical Skills
- Strong Python — our primary language across AI and backend
- APIs: RESTful services, webhooks, third-party integrations; FastAPI or Django preferred
- Databases: PostgreSQL — schema design, indexing, query performance; Redis or similar
- Async Python (asyncio) and event-driven architectures
- Cloud: working knowledge of AWS (or GCP/Azure) — compute, storage, managed DBs, queues
- Version control, code review, and CI/CD as normal parts of your workflow
AI Domain Understanding
- Shipped an LLM system to real users and debugged it in production; can describe a failure mode you found and fixed
- Specific view of what current models are unreliable at, and how that shapes what runs unsupervised versus behind a tool call or human review
- Multi-turn conversation design — state that survives interruptions, topic switches, and mid-call corrections
- Experience measuring LLM systems: offline evals, judge model calibration, or production quality metrics
- Working knowledge of token cost and latency trade-offs, including where streaming helps and where it adds complexity
Soft Skills
- Frame the problem, not just solve it; write down architectural decisions and the tradeoffs you rejected
- Clear communication with technical and non-technical stakeholders
- Able to drive your own work to completion without close supervision
- Comfortable with a fast pace and evolving requirements
- Willing to work in-office in Gurugram and overlap with US hours when needed
Preferred Experience & Skills
- Conversational AI — chatbots, voice assistants, or IVR
- Voice/telephony (Twilio, Vonage) or STT/TTS APIs (Deepgram, ElevenLabs, AssemblyAI)
- LLM orchestration frameworks (LangChain, LlamaIndex) — or a considered view on skippingthem
- Healthcare / HIPAA compliance knowledge
- SaaS or B2B product company background; multi-tenant architecture