I'm an AI/ML Engineer at SunRay Construction Solutions in Hyderabad. Before I built AI for construction payment protection, I spent years doing the work myself, as a research analyst and then a data scientist. That work covered Notice to Owner research, lien and payment-rights deadlines, and reconciling what a customer claims against what the documents actually show.
That background shapes how I build. I know where ambiguity, missing evidence and compliance risk enter a workflow, so my systems are designed around them:
- Evidence before answers. Agents cite source records, keep what the customer said separate from what was verified, and never overwrite the original data.
- Probabilistic AI, deterministic control. Models produce structured findings. Plain application code decides what happens next, so model output never directly changes an operational state.
- Escalate, don't guess. When documents conflict, the system routes the case to a human with the discrepancy spelled out, rather than inventing a resolution.
- Honest evaluation. Labelled synthetic datasets, frozen versions, and mistakes corrected in the open rather than hidden.
| Project | What it is | Architecture highlights | Stack |
|---|---|---|---|
| construction-legal-ai-suite | AI for construction payment protection: work-order intelligence, Notice to Owner research coaching and payment-risk prediction | Microsoft Foundry specialist agents · MCP retrieval tool on AWS Lambda · deterministic Python router · bounded correction loop · human review | Python · FastAPI · MCP · AWS · Azure · Docker · Next.js |
| portfolio | Source for gopalakrishnagenai.in: case studies, verified credentials and an integrated AI assistant | Streaming AI assistant grounded in portfolio content · server-side API routes · SEO metadata and JSON-LD | Next.js · TypeScript · Tailwind · Vercel |
| FrontierWeekHack | My working copy of the Microsoft Foundry Frontier Week hackathon labs | Building, tracing, evaluating and deploying agentic workflows on Microsoft Foundry | Microsoft Foundry · Python · Azure |
Measured on the synthetic evaluation sets published in the repositories:
- 154 / 156 (≈99%) on the work-order intake-agent evaluation, with 100% tool selection, tool-call success and tool-call accuracy
- 120 work orders across 24 scenario families in the labelled evaluation corpus (versioned; the earlier version is frozen)
- 75 automated tests passing on the multi-agent control plane: router, parser, correction overlay, observability and API
- 0.684 holdout ROC-AUC for the best AutoML challenger on payment-risk prediction, benchmarked against an interpretable Logistic Regression production model
- Two live Azure deployments (work-order operations console and NTO research coach), shipped through GitHub Actions with OIDC
- AI and ML: Python · PyTorch · scikit-learn · LangChain · LangGraph · RAG · vector databases · SHAP · DataRobot
- Agents: Microsoft Foundry · Model Context Protocol · multi-agent orchestration · evaluation and tracing
- Engineering: FastAPI · Pydantic · Next.js · TypeScript · PostgreSQL · Docker
- Cloud and MLOps: AWS (S3, Glue, Athena, SageMaker, Lambda, API Gateway) · Azure (Container Apps, App Service, ACR) · GitHub Actions
- Post Graduate Program in AI and Machine Learning: McCombs School of Business, The University of Texas at Austin (2021)
- Founderz Agent Explorer and Agent Architect (2026)
- IBM Generative AI specializations for data scientists and data analysts, plus MLOps and Practical Data Science specializations from DeepLearning.AI
- See all credentials →
Open to Generative AI Engineering, Agentic AI and ML Engineering roles.