Deputy Unit Manager - Finserv Intelligence

Location Name: Pune Corporate Office - Mantri

Job Purpose

As a Research Scientist in the BFS Intelligence Division, you will lead the charge in developing Sovereign AI — models designed specifically for the linguistic, cultural, and structural complexities of the Indian subcontinent. You are not merely fine-tuning open-source models; you are architecting the Small Language Models (SLMs) and Multimodal Engines (Vision, Voice, Agentic AI) that will power the next decade of Indian finance, retail, and digital public infrastructure.

You will operate in a 'Dual-Helix' environment: 50% deep-tech discovery in IIT Research Labs and 50% high-stakes industrialization within our Corporate Innovation Hub, bridging academic theory with enterprise-scale deployment at a pace that sets India's AI agenda globally.

Duties And Responsibilities

Minimum Required Accountabilities for this Role

Frontier R&D: The SLM & Multimodal Roadmap Sovereign SLMs: Architect and pre-train domain-specific Small Language Models (1B–7B parameters) that outperform GPT-4 in niche BFSI tasks while running on cost-effective, edge-based hardware. Multimodal Fusion: Develop 'Vision-Voice-Text' integrated models for hyper-personalized customer experience — e.g., AI agents that can 'see' a customer's physical document via camera and 'talk' through a complex loan process in regional dialects. Geospatial & Alternative Intelligence: Build models that fuse satellite imagery and geospatial data with financial transaction patterns to predict credit risk in rural and semi-urban markets. Data Engineering & Ethics Synthetic Data Synthesis: Create high-fidelity, privacy-preserving synthetic datasets to train models where real-world BFS data is restricted or sparse. Agentic Frameworks: Research 'Chain-of-Thought' and 'Reasoning-Action' (ReAct) patterns to transform chatbots into Autonomous Financial Agents capable of executing cross-platform transactions. IIT Ecosystem Leadership Academic Bridge: Lead collaborative 'Moonshot' projects with IIT faculty, ensuring that theoretical breakthroughs in neural architecture search or attention mechanisms are patented and productized. Knowledge Sovereignty: Represent the firm at global forums (NeurIPS, ICML, CVPR) to showcase 'India-First' AI innovations, moving the needle from consumption to creation.

Additional Accountabilities Pertaining To This Role

Mentor B.Tech/M.Tech interns from IIT partner institutions, building a pipeline of Sovereign AI talent. Serve as Technical Advisor to portfolio startups within the $50M Deep-Tech Portfolio. Drive IP strategy — identify patentable innovations across SLM architecture, multimodal fusion, and agentic frameworks. Collaborate with Consumer, Retail, Digital Platforms, Customer Experience, and Enterprise Process teams to convert research into production AI solutions.

Key Decisions / Dimensions

Independent Decisions

Research methodology, model architecture selection, and experimental design for SLM and multimodal projects. Synthetic dataset generation strategies and evaluation benchmarks. Identification of patentable innovations and priority filing recommendations. Intern project scoping and day-to-day research mentorship.

Decisions Requiring Approval

Productionization and deployment of AI models into live BFSI systems. Engagement of external academic collaborators and joint publication agreements. Budget allocation for compute (GPU clusters) and dataset procurement. Representation at international conferences and sabbatical program participation.|Financial Dimensions Business Volume: Indirect impact across the BFSI digital ecosystem powered by SLM and Multimodal AI outputs at scale. Venture Exposure: Direct access to the $50M Deep-Tech Portfolio with opportunities to serve as Technical Advisor to funded AI startups. Commercialization: Tiered incentive/bonus for models that are productionized and scaled; separate track for IP filings and licensed patents.

Other Dimensions

Work Model: Dual-Helix — 50% IIT Research Labs (deep-tech discovery), 50% Corporate Innovation Hub (industrialization). Rotation Model: Roster-based presence across IIT campuses (Delhi, Mumbai, Madras, Kharagpur) and Metro Centers. Deep-Tech Sabbaticals: Sponsored residencies at partner IIT labs for independent research aligned with the firm's long-term vision. Number of Direct Reports: As per Research Cluster expansion trajectory. Number of Locations: Multiple — Corporate Innovation Hub + IIT campuses across India.

Major Challenges

Pre-training and fine-tuning 1B–7B SLMs for BFSI-specific tasks with limited labeled Indian-language data. Achieving real-time multimodal inference (Vision-Voice-Text) at the latency standards required for live customer interactions and Video-KYC workflows. Building privacy-preserving synthetic datasets that faithfully replicate the statistical properties of restricted real-world BFS data. Translating ReAct/agentic reasoning research into production-safe Autonomous Financial Agents that comply with RBI and SEBI regulatory frameworks. Maintaining academic publication velocity while simultaneously industrializing models inside a high-stakes corporate environment. Balancing 50% IIT lab discovery time against 50% Corporate Innovation Hub delivery commitments across a multi-campus rotation model.

Educational Qualifications

Required Qualifications and Experience

PhD or Post-Doctoral qualification from a top-tier global or Indian institution (IIT, IISc, CMU, Stanford), specializing in Deep Learning, NLP, or Computer Vision — preferred. M.Tech from an equivalent Tier-1 institution also considered.

Work Experience & Core Competencies

The 'Build' Mindset: Expert-level proficiency in PyTorch and/or JAX; experience with high-performance computing (CUDA, Triton) and distributed training across GPU clusters. Innovation Track Record: Portfolio of peer-reviewed publications OR a GitHub repository demonstrating implementation of frontier architectures (Transformers, State Space Models/Mamba, Diffusion Models). Strategic Fungibility: Ability to pivot between Speech AI (ASR/TTS for 22+ Indian languages) and Transactional AI (fraud detection, risk modeling) as cluster priorities evolve. India Stack Fluency: Deep understanding of India's digital public infrastructure (UPI, ONDC, Account Aggregator, ABDM) and its AI opportunity surface. Agentic AI Experience: Familiarity with Chain-of-Thought, ReAct, and multi-agent orchestration frameworks for building Autonomous Financial Agents. Communication & Collaboration: Demonstrated ability to publish, present, and translate research into enterprise-grade PoCs. Experience mentoring IIT interns is an added advantage.