Job Purpose

Design, test, and productionize high performing prompts and orchestration patterns for Large Language Models (LLMs) and Vision Language Models (VLMs) that augment Indus Towers' Computer Vision pipelines (OCR, defect detection, PTW/ safety compliance, field audits, Preventive Maintenance Audits etc.). The role ensures accuracy, reliability, cost efficiency, and latency targets are met in production by applying state of the art prompting and rigorous evaluation.

Principal Accountabilities

1. Production grade prompting & orchestration

- Ship robust prompt chains using few shot, chain of thought (CoT), self consistency CoT, least to most, and variants (e.g., debate/Tree of Thought).
- Implement guardrails and structured output (JSON) contracts; integrate retries, validators, and fallbacks.

2. LLM sampling strategy design & tuning

- Tune temperature, top k/top p (nucleus), beam search, repetition penalty, max tokens, and multi pass/self consistency sampling.

3. Error analysis & mitigation

- Identify and mitigate hallucinations, instruction drift, context leakage, prompt fragility, and format non compliance.

4. Evaluation & experimentation

- Set up prompt evaluation suites with metrics: accuracy, pass@k, win rate vs. baselines, faithfulness, format compliance, toxicity/safety, latency, and cost.

5. Tools & platform integration





- Build prompt pipelines in Python; integrate with Lang Chain, Pinecone, and prompt evaluation tools.

6. Domain collaboration & requirements mapping

- Work closely with Product and Business teams to translate tower operations rules into effective prompt policies.

7. Research tracking & continuous improvement

- Stay current with state of the art prompting and LLM/VLM research; introduce relevant techniques to improve robustness and reduce cost.

Required Qualifications & Skills

- 3-4 years of production experience designing advanced prompts: few shot, CoT, self consistency CoT, least to most.
- In depth knowledge of LLM sampling strategies and trade offs.
- LLM error taxonomy & mitigation strategies.
- Up to date with latest research and state of the art prompting techniques.
- Proficient in Python for rapid prompt iteration.
- Hands on with Lang Chain, Pinecone, and prompt evaluation tools.
- Strong debugging skills.
- Collaboration and communication skills.

Preferred / Good to Have

- Experience with Vision Language Models and OCR+LLM pipelines.
- Familiarity with guardrails & safety tooling.
- Exposure to Azure/OpenAI/Hugging Face ecosystems.
- Background in Computer Vision (OpenCV, YOLOv8, segmentation, OCR).
- Experience with A/B testing and statistical analysis.

📌 Prompt Engineering (India)
🏢 Objectwin Technology India
📍 India