AI Agentic Engineer – Associate Software Engineer at Talentum (2025-10 – Present)
Tools: LangChain | AutoGen | Coyote Analytic Software | SQL | XML | Python | LLM APIs (Claude, OpenAI)
- Design and build LLM-powered agentic workflows that automate report drafting, validation, and QA on top of the existing SQL-based reporting pipeline, reducing manual review steps for operational, financial, and client-specific reports.
- Develop autonomous agents that draft executive summaries, flag data inconsistencies, and validate report output against source data before it reaches clients.
- Integrate Claude and OpenAI APIs into agent pipelines to automate data validation and anomaly detection across recurring reporting workflows.
- Work directly with clients and business stakeholders to gather reporting requirements and translate them into agent-driven automation and scalable reporting solutions.
- Maintain XML invoice and check formats, incorporating agent-based validation checkpoints into client-specific billing and compliance workflows.
- Study relational database structures and reporting schemas to inform agent design, and troubleshoot report generation issues, SQL errors, and XML formatting defects in production.
- Partner with Operations, Product, and Client Support on agent and report enhancements; document agent architecture and lead knowledge-transfer sessions for the team.
Automation & AI Engineer Intern at Cyient Ltd. (2025-06 – 2025-09)
Hybrid
- Deployed 4 deep learning pipelines (a CNN image classifier at 92% accuracy, plus transformer NLP models) across AWS SageMaker, Azure ML, and GCP Vertex AI, and kept 99.5% production availability over a 4-month tenure.
- Set up OCR-based document processing for 1,000+ scanned files, cutting manual data entry by half, and led 5 ethical AI compliance reviews that sped up review cycles by 20%.
- Tracked 10+ ML models in MLflow, which raised predictive accuracy by 18%, lowered false positives by 12%, and reduced retraining time by 25%.
Freelancer at Freelancer (2023-11 – 2025-05)
Focus: RAG Systems | Multi-Agent Orchestration | Report Automation
- Delivered 5 end-to-end LLM projects covering RAG pipelines, multi-agent orchestration, prompt evaluation, and LLM observability, each containerized with Docker and deployed via FastAPI.
- Increased RAG pipeline relevance by 45% (FAISS, ChromaDB, Qdrant, Pinecone) using semantic and hybrid retrieval, confirmed through LangSmith regression suites on held-out test sets.
- Created PromptGuard, a prompt evaluation framework tested on 500+ adversarial edge cases, which shortened new-feature validation time from 3 weeks to 4 days and raised output reliability by 45%.
- Designed AutoTasker, an AutoGen multi-agent system (Planner / Executor / Critic) with Redis-backed memory, cutting manual task steps by 60% across 200+ workflow runs while keeping P95 latency under 4 seconds.
