We are hiring for 4 - AI/ML Lead positions as below. When applying, please select only ONE specialisation that best matches your expertise and experience.

1. NLP & Document AI

OCR and document parsing, information extraction, PII redaction, and retrieval-augmented question answering

2. Speech AI

Speech recognition and synthesis for Indian languages, and voice and chat assistants

3. Computer Vision

Face verification and liveness, document forgery detection, and image and geospatial analytics

4. Agentic AI

LLM agents that complete multi-step work in enterprise systems under human approval.

Role Summary

The AI/ML Lead will lead an AI team in one specialisation, owning the design, development, evaluation and deployment of reusable AI services and providing technical oversight of vendor teams. The role requires hands-on depth in the specialisation, rigorous evaluation, and the judgement to build services once for reuse across government departments, using open-weight and Indian models hosted on IndiaAI compute.

Educational Qualification

- B.Tech./B.E., M.Tech./M.S. or M.Sc. in Computer Science, AI/ML, Data Science, Electronics or a related discipline (Must have)
- Advanced degree (M.Tech./M.S./Ph.D.) or published work in the chosen specialisation is highly desirable
- Certifications are not mandatory; demonstrated work, as listed under Minimum Evidence at Screening, shall carry greater weight
- Candidates from non-traditional backgrounds with solid verifiable work will be considered in lieu of formal qualification

Experience

- 7+ years developing and deploying AI/ML systems (10+ years preferred); minimum 4 years building systems in the chosen specialisation deployed in production (Agentic AI: minimum 2 years building LLM applications or agents in production)
- At least 2 years leading a team of engineers or data scientists, including setting technical direction and reviewing their work
- Demonstrable delivery of at least one system in the chosen specialisation used by real users at scale, with documented evaluation: field-level accuracy (NLP & Document AI), word error rate or resolution rate (Speech AI), error rates at operating thresholds (Computer Vision), or task-success rates and safety controls (Agentic AI)
- Hands-on experience fine-tuning open-weight or foundation models on domain data




- Experience with Indian languages, scripts or code-mixed data is a strong plus
- Experience overseeing vendor delivery is desirable

Key Responsibilities

Technical Leadership

- Own the technical roadmap of the team's reusable services, choosing between prompting, retrieval, fine-tuning and smaller specialised models on accuracy, cost, latency and data-governance grounds, and documenting the rationale
- Lead in-house engineers and data scientists, review their work, and remain hands-on in code, training and debugging

Specialisation Responsibilities

- NLP & Document AI: OCR, layout analysis and parsing for scanned, handwritten and mixed-script documents; extraction, classification, PII redaction, summarisation, similarity and de-duplication; retrieval-augmented question answering over rules, circulars and case files, with citations and grounding checks
- Speech AI: speech recognition, synthesis, translation and language identification for Indian languages, including 8 kHz telephone audio and code-mixed speech; assistants on web, WhatsApp, IVR and mobile with retrieval-grounded answers, human hand-off and safe failure; real-time voice pipelines within latency budgets
- Computer Vision: face detection, 1:1 verification, 1:N search, liveness and presentation-attack detection at large scale; tamper, forgery and photo-substitution detection for identity documents and certificates; field-photo and satellite imagery analytics; deployment on server, mobile and edge
- Agentic AI: agents for planning, tool use, memory and multi-agent coordination using LangGraph or equivalent and the Model Context Protocol (MCP); least-privilege tool permissions, human approval for every consequential action, audit trails and rollback; reusable agent templates for drafting, document scrutiny and request routing

Evaluation & Safety

- With the QA Lead, build golden datasets and specialisation-appropriate metrics, set release thresholds,



and run subgroup analysis across languages, document types, demographics or acoustic conditions
- With the AI Safety team, test services for prompt injection, presentation attacks, adversarial inputs and leakage of personal data before release

Deployment & Reuse

- Deploy services through the AI platform and gateway, publish them to the service catalogue with documentation and SLAs, and contribute models and evaluation sets to AIKosh

Vendor Oversight

- Review designs, code and evaluation evidence of vendor teams, and recommend acceptance or rejection of deliverables

Responsible AI & Privacy

- Apply anonymisation, data minimisation and DPDPA 2023 requirements, with stricter handling for biometric data in Computer Vision, and document models through model cards and evaluation records

Technical Competencies

- Common: Python (expert level), PyTorch, Hugging Face Transformers, PEFT and TRL (LoRA, QLoRA, DPO, GRPO); vLLM or SGLang; MLflow, Docker, Git; evaluation design with bootstrap confidence intervals; Langfuse or Arize Phoenix
- Indic Language AI: AI4Bharat models (IndicConformer, IndicTrans2, IndicBERT, Indic Parler-TTS), Sarvam models and Bhashini services; code-mixed and transliterated text
- NLP & Document AI: document-parsing models such as PaddleOCR-VL, MinerU or GLM-OCR, and PP-OCR; Docling; vision-language models such as the Qwen-VL family; hybrid search and re-ranking; pgvector, Qdrant or Milvus; RAGAS or DeepEval; Microsoft Presidio
- Speech AI: IndicConformer, Whisper, NVIDIA Parakeet or Canary; NVIDIA NeMo or ESPnet; Silero VAD, pyannote; WebRTC, SIP, Pipecat or LiveKit Agents; WhatsApp Business Platform; WER and CER tooling
- Computer Vision: YOLO26 or YOLO11, RF-DETR, SAM 3, DINOv3, SigLIP 2; ArcFace or AdaFace with InsightFace; ISO/IEC 30107-3 presentation-attack metrics; image-manipulation detection (such as TruFor); ONNX Runtime, TensorRT, LiteRT
- Agentic AI: LangGraph, LlamaIndex, OpenAI Agents SDK or Google ADK; MCP; Temporal; Open Policy Agent; trajectory and tool-use evaluation (Inspect, DeepEval or LangSmith); prompt-injection testing for agents (such as AgentDojo)
- Governance: DPDPA 2023, MeitY Responsible AI advisory, IndiaAI Safe & Trusted AI framework, OWASP Top 10 for LLM Applications

📌 AI/ML Lead (Noida)
🏢 National e-Governance Division (NeGD), Digital India
📍 Noida