Director Data Science

ABOUT CREDGENICS

Credgenics is India's leading SaaS platform for debt collections and resolution, working with banks, NBFCs, ARCs, and

fintech lenders. Our platform spans digital communications, field collections, legal workflows, payments, and AI-driven

collections strategy — now expanding into multilingual AI voice bots for large-scale collection conversations.

THE ROLE

We are hiring a Head of Data Science to own two mandates:

(1) Applied AI for our Voice Bot — the single most important product bet at Credgenics today, and (2) the core Data Science charter powering collections strategy across the platform.

This is a leadership role — you will set strategy, build the team, and drive technical direction. You must be technically deep enough to review evals, challenge model choices, and make build-vs-buy decisions with conviction.

WHAT YOU WILL OWN

Voice AI — Flagship Mandate

• Own end-to-end ML quality of the voice bot pipeline: STT accuracy on noisy Indian-language telephonic audio,

TTS naturalness, LLM conversation quality, and full-loop latency (VAD, endpointing, streaming)

• Build rigorous offline and online eval frameworks — disposition accuracy, script & regulatory compliance,

hallucination/safety red-teaming, containment, and conversion (PTPs, payments). Every model change ships

against an eval

• Make vendor and architecture decisions: STT/TTS/LLM selection, fine-tuning vs. prompting, self-hosted vs. API,

and data residency for Indian BFSI clients

• Build the data flywheel: turn call recordings, transcripts, dispositions, and QC audits into training data and

continuous improvement loops

• Partner with Product and Engineering on conversation design — objection handling, drop-off analysis, call-

outcome modelling, and per-client tuning

• Building and training in-house STT and TTS models in the longer run.

Collections Data Science

• Own and improve the predictive model suite: propensity to pay, best time/channel to contact, agency allocation,

settlement & legal-action propensity, roll-rate and recovery forecasting

• Build an experimentation culture: champion/challenger testing, A/B frameworks, and uplift measurement that

clients trust

• Package model outputs and portfolio insights as product features, not one-off decks

• Own model monitoring, drift detection, retraining pipelines, and audit-ready documentation

Leadership & Governance

• Hire, grow, and retain a high-performing team of data scientists, ML engineers, and applied AI engineers

• Set the DS/AI roadmap jointly with Product and Engineering; present outcomes and trade-offs to founders and client CXOs

• Own model governance: explainability, bias/fairness checks, auditability, and compliance with RB expectations and India's DPDP Act

• Establish MLOps foundations: experiment tracking, model registry, versioning, monitoring, and rollback

WHAT WE ARE LOOKING FOR

Must Have

• 10+ years in data science / applied ML, with 4+ years leading and scaling teams

• Demonstrated production experience with LLM-based or speech systems in the last 2 years — you have shipped conversational AI, GenAI apps, or speech pipelines at real scale

• Strong classical ML foundations: supervised learning, scorecards/propensity models, experimentation, and causal measurement

• Experience deploying models in regulated environments (BFSI strongly preferred; lending/collections a major plus)

• Ability to make architecture and vendor decisions under commercial constraints — you think in unit economics, not just accuracy metrics

• Clear communication with engineers and business leaders; comfortable presenting to clients.

Strong Plus

• Hands-on exposure to the modern voice stack: STT (Deepgram, Sarvam, Whisper), TTS (ElevenLabs, Azure), real-time orchestration (Pipecat, LiveKit), telephony integration

• Multilingual and vernacular NLP experience — code-switched conversations, regional accents, and ASR/TTS quality on Indian-language telephonic audio

• Experience with LLM eval tooling, red-teaming, and guardrail design

• Prior experience in lending, collections, or contact-centre AI

WHAT SUCCESS LOOKS LIKE (FIRST 12 MONTHS)

• Voice bot quality metrics (containment, disposition accuracy, compliance, PTP conversion) are defined, instrumented, and improving MoM with a rigorous eval pipeline behind every release.

• Vendor stack decisions (STT/TTS/LLM, hosting) are settled with clear cost and quality rationale

• Core collections models are refreshed, monitored, and demonstrably lifting client recovery metrics

• A strong DS/Applied AI team is in place with clear ownership and low regrettable attrition

• Banking clients trust our AI: governance documentation, explainability, and audit readiness are in place