AI/ML Engineer building production agentic systems, RAG pipelines, and explainable ML, wherever the domain takes me.
I build AI systems that reason, retrieve, and act, then hold up in production. My day job is healthcare AI; my own projects are mostly fraud, risk, and telecom. I like it that way; the same core skills (agent orchestration, RAG, explainable ML, evaluation) transfer cleanly across domains, and switching between them keeps me honest about what's actually general-purpose versus what's just familiar.
- 🏥 At work: AI Engineer at CareCloud (NASDAQ: CCLD), building multi-agent LLM pipelines that turn clinical documents into structured decisions, at 10,000+ documents/day.
- 💳 On my own time: fraud detection, telecom churn, and RAG+SQL agents, built to prove the same engineering discipline holds outside healthcare.
- 🎬 Also shipping: a data lake and Power BI monitoring for an AI captioning product, and a computer-vision pipeline underneath it.
Promoted from AI Intern to AI Engineer in under a year by shipping systems, not demos.
AI Engineer, CareCloud (NASDAQ: CCLD) · May 2025 – Present Multi-agent LLM pipelines (LangGraph), fine-tuning (Qwen 14B, LoRA), and a 283K-sample evaluation harness for medical coding, served on vLLM + FastAPI.
AI/ML Engineer, Kalakar (Part-time) · Feb 2025 – Present Data lake for multilingual captioning model training, plus a Power BI dashboard tracking pipeline success rate.
AI Engineer, R&D, DevHawks · Nov 2024 – Apr 2025 Benchmarked 20+ LLMs across domain tasks and reviewed 20+ SOTA papers on causal AI, agents, and explainable AI; built reusable evaluation pipelines cutting experimentation time ~40%.
AI / LLM
Backend & Data
ML, Explainability & BI
Tools
💳 ledger-ai · Python, DuckDB, Scikit-learn, SHAP, FastAPI Fraud investigation engine: leak-safe SQL feature pipeline over 24.4M transactions, hybrid Isolation Forest + Random Forest at 0.97+ PR-AUC, tiered risk routing catching ~98% of fraud at 75-80% review precision.
📡 insight · Python, LangChain, FAISS, SQLite, Gemini Telecom retention copilot: routes queries between a safe SQL tool and RAG retrieval with zero cross-contamination, enforcing strict numeric grounding so the model never cites a number it wasn't given.
🧭 huma · Python, Ollama, SQLite, Typer CLI agent that fetches job listings from four boards and scores each against your resume using a fully local LLM; deterministic FSM pipeline, nothing leaves your machine.
📉 commercial_subscriber_risk_model · Python, LightGBM, SHAP Telecom churn intelligence: K-Means behavioral segmentation plus a LightGBM classifier (0.81 ROC-AUC), threshold-calibrated for a retention-first strategy.
🏦 Vera · Python, Streamlit, Scikit-learn Loan approval prediction and analytics: single and batch decisions with confidence scores, an explainability view, and synthetic data generation.
🎥 video_pipeline · Python, FastAPI, Docker, SAM2 Prompt-driven video subject/background separation: Grounding DINO detection + SAM2 tracking, served as an async FastAPI job service.
🏥 Aura · FastAPI, Node.js, LightGBM, SHAP (repo coming soon) Full-stack clinical decision support platform for 30-day hospital readmission risk.
Multi-agent orchestration · Model Context Protocol (MCP) · LLM evaluation and hallucination reduction · RAG grounding and retrieval quality · explainable ML for high-stakes decisions
LinkedIn · tikram179@gmail.com · GitHub