AI/ML Lead for Org
Budget: ₹37,500 – ₹75,000 INR
The project brings together computer vision, natural-language processing, and traditional machine-learning techniques to power a new deep-tech product that will move from prototype to production. I need end-to-end technical leadership that covers strategy, experimentation, and hands-on model delivery.
Current status
• Early-stage proof-of-concept code bases exist for both vision and text tasks, but they live in silos.
• No unified data pipeline, no MLOps layer, and no clear roadmap tying research spikes to shippable features.
Scope of work
• Audit the existing repos, datasets, and annotation processes, then define a clean, version-controlled baseline.
• Design an integrated architecture that supports multi-modal learning (images + text).
• Select or fine-tune state-of-the-art models—e.g., ViTs, CLIP, BERT, or transformer hybrids—balancing accuracy with inference speed.
• Set up automated training, testing, and deployment workflows using tools such as TensorFlow/PyTorch, Docker, and a CI/CD stack (GitHub Actions, Kubeflow, or similar).
• Establish data governance, monitoring, and retraining triggers so the system stays reliable in production.
• Produce concise technical documentation and a knowledge-transfer session for the in-house dev team.
Deliverables
1. Architecture diagram and technical design document.
2. Re-factored, container-ready codebase with reproducible builds.
3. Trained models with evaluation reports against agreed metrics.
4. CI/CD and MLOps scripts deployed in a staging environment.
5. Walk-through video or live hand-over meeting plus written docs.
Acceptance criteria
• End-to-end pipeline trains and deploys with a single command.
• Model performance meets or exceeds the current PoC baselines on both image and text tasks.
• Deployment container passes load testing at the target QPS and latency.
• All code and assets are version-controlled and documented.
Timeline and collaboration
I’m aiming for an initial production release in 10–12 weeks, with weekly milestone check-ins. Direct Slack access will be provided for rapid feedback, and any tooling preferences that accelerate delivery can be accommodated.
Current status
• Early-stage proof-of-concept code bases exist for both vision and text tasks, but they live in silos.
• No unified data pipeline, no MLOps layer, and no clear roadmap tying research spikes to shippable features.
Scope of work
• Audit the existing repos, datasets, and annotation processes, then define a clean, version-controlled baseline.
• Design an integrated architecture that supports multi-modal learning (images + text).
• Select or fine-tune state-of-the-art models—e.g., ViTs, CLIP, BERT, or transformer hybrids—balancing accuracy with inference speed.
• Set up automated training, testing, and deployment workflows using tools such as TensorFlow/PyTorch, Docker, and a CI/CD stack (GitHub Actions, Kubeflow, or similar).
• Establish data governance, monitoring, and retraining triggers so the system stays reliable in production.
• Produce concise technical documentation and a knowledge-transfer session for the in-house dev team.
Deliverables
1. Architecture diagram and technical design document.
2. Re-factored, container-ready codebase with reproducible builds.
3. Trained models with evaluation reports against agreed metrics.
4. CI/CD and MLOps scripts deployed in a staging environment.
5. Walk-through video or live hand-over meeting plus written docs.
Acceptance criteria
• End-to-end pipeline trains and deploys with a single command.
• Model performance meets or exceeds the current PoC baselines on both image and text tasks.
• Deployment container passes load testing at the target QPS and latency.
• All code and assets are version-controlled and documented.
Timeline and collaboration
I’m aiming for an initial production release in 10–12 weeks, with weekly milestone check-ins. Direct Slack access will be provided for rapid feedback, and any tooling preferences that accelerate delivery can be accommodated.