Trayan Das

Forensic Knowledge Assistant (Multimodal RAG): Designed a citation-grounded Q&A assistant (Microsoft Foundry, Azure AI Search, Azure Functions, Copilot Studio, Power Platform) answering engineering-investigation questions from text, an image, or both; enforced grounding in code under a cite-or-silent policy, secured with Entra ID; recall@10 0.84, RAGAS factual correctness 0.87, 2% false-answer / 5% false-refusal, p95 8s text-only / 20s with image.

AI/ML Computational Science Specialist at Accenture (2026-09 – Present)

• Forensic Knowledge Assistant (Multimodal RAG): Designed a citation-grounded Q&A assistant (Microsoft Foundry, Azure AI Search, Azure Functions, Copilot Studio, Power Platform) answering engineering-investigation questions from text, an image, or both; enforced grounding in code under a cite-or-silent policy, secured with Entra ID; recall@10 0.84, RAGAS factual correctness 0.87, 2% false-answer / 5% false-refusal, p95 8s text-only / 20s with image.

AI/ML Computational Science Sr Analyst at Accenture (2024-09 – 2026-08)

• Privacy-Safe Analytics Agent: Architected an Azure analytics platform — incremental ingestion with PII/profanity redaction, hybrid BM25 + vector AI Search over 4M+ source records consolidated into 100K+ indexed documents, four-agent LangGraph reasoning layer — shipped via Docker/GitHub Actions CI/CD with OpenTelemetry and Application Insights tracing; routing accuracy 97%, query-parsing 81%, retrieval precision 0.91 / recall 0.92; cutting time-to-insight from hours to seconds (~95% faster), retrieval latency ~75%, analyst effort 60–70%.
• Enterprise Data Q&A Agent (MCP-Grounded Text-to-SQL): Engineered an MCP-grounded text-to-SQL system — MCP servers on Azure Container Apps exposing Fabric Lakehouse schema as resources and hardened execution (SELECT-only validation, row caps, self-correction loop) as tools, letting a single LangChain query agent hold several dataset schemas at once for verified cross-dataset joins, behind a Copilot Studio orchestrator. Per-user row-level security via on-behalf-of Entra ID token exchange; 10M+ records in <30s, routing accuracy 94% on 148 questions.
• GenAI Business Doc Generator: Delivered a multi-agent RAG system with LangGraph + LoRA fine-tuned LLM for business document generation; LangSmith tracing, ROUGE evals; cutting drafting effort 70–80% at 100+ documents/week.

Software Developer at Expro (2022-09 – 2024-09)

• Well Performance Analytics Platform: Implemented a PyTorch LSTM-autoencoder pipeline for anomaly detection on well-sensor time-series data, improving risk detection by ~20% and reducing analysis time by ~40%.
• Oil & Gas Insights Web Application: Designed a PySpark KPI pipeline on Azure Databricks using PCA, K-means, and Random Forest to derive KPI features from equipment telemetry, reducing manual reporting by ~50%.

Data Scientist at Ganit Labs (2021-03 – 2022-09)

• Equipment Failure Prediction Model: Developed predictive maintenance models on GCP for oilfield equipment with ~85% recall and ~30% less downtime.

Biomedical Engineer at Portea, AOV, Care Group (2018-08 – 2020-09)

• Led cross-functional teams in medical equipment maintenance and training programs.

MTech (Master of Technology) – University of Calcutta (2016 – 2018)

BTech (Bachelor of Technology) – WBUT (2011 – 2015)