Prakash Nikam

AI/ML Engineer with 4 years of industry experience building production grade Generative AI and machine learning systems for real world AI applications. Skilled in developing agentic AI workflows, RAG pipelines, and scalable AI services using LangGraph, LangChain, and Vertex AI. Hands-on experience in end-to-end AI system development including model development, retrieval optimization, FastAPI services, and cloud deployment on GCP and AWS.

AI/ML Engineer - Codify Software Services - GCP

(2025-06)

Agentic AI Customer Support Automation System

• Built a multi-agent AI workflow using LangGraph and Gemini via Vertex AI to automate customer support tasks including refunds, order tracking, and query handling through intent-based routing.
• Designed a RAG pipeline using LangChain and Pinecone, improving retrieval relevance through chunking optimization, overlap tuning, and prompt engineering
• Developed intent classification and agent-routing workflows to manage retrieval, API execution, and fallback handling
• Reduced LLM inference cost by ~40% through model routing between Gemini Flash for lightweight queries and Gemini Pro for complex multi-step reasoning tasks
• Deployed scalable FastAPI services on Cloud Run with asynchronous request handling and automated CI/CD pipelines using Cloud Build

AI/ML Engineer - PHN Technology - AWS

(2024-06 - 2025-06)

Multilingual RAG GenAI EdTech Assistant

• Built a multilingual RAG assistant supporting educational queries across Hindi, English, Marathi, and Telugu using LangChain, Sentence Transformers, and Pinecone
• Implemented hybrid retrieval using BM25, semantic search, and RRF fusion with reranking, improving RAGAS answer relevance from 0.61 to 0.82
• Reduced end-to-end response latency by ~50% using Redis semantic caching and async retrieval pipelines
• Engineered modular AI pipelines with retrieval, reranking, guardrails, and fallback orchestration; tracked experiments and retrieval optimization workflows using MLflow
• Containerized and deployed AI services using Docker and AWS EC2 for scalable inference and reproducible experimentation

AI/ML Engineer - PHN Technology

(2024-06 - 2025-06)

ASTER Humanoid Robot AI-Based Embedded System

• Built a real-time pose-mimicking system using MediaPipe Holistic and Raspberry Pi 4B, achieving stable tracking at 20–25 FPS
• Applied Kalman filtering on 33 body landmark coordinates for smooth servo motor movement, eliminating chatter from frame-to-frame jitter in joint angle estimates
• Optimized inference using ONNX Runtime for faster CPU execution and improved real-time robotics performance
• Integrated offline Speech-to-Text and NLP-based voice + gesture interaction for multimodal robot control

ML Engineer - Codify Software Services - AWS

(2023-04 - 2024-03)

Bank Customer Churn Prediction

• Trained churn prediction models using Random Forest, Logistic Regression, and LinearSVC on large scale banking data (~2M records), achieving ROC-AUC of ~0.85
• Engineered customer behavioural features including RFM patterns (recency, frequency, monetary), tenure, digital engagement score, and complaint frequency to improve prediction performance
• Applied SMOTE and hyperparameter tuning to improve model stability and generalization.
• Extracted customer data from banking databases using SQL pipelines and stored processed datasets in S3 for ML training workflows.
• Built FastAPI inference APIs and deployed prediction services using Docker and AWS Lambda.
• Used MLflow for experiment tracking, model versioning, and performance comparison across training runs.

Graduate Research Trainee - SAMEER — IIT Bombay

(2019-02 - 2020-02)

Retinal OCT Image Analysis Medical Imaging Research

• Assisted in a 10-year OCT research program; collected raw spectrometer signals from SD-OCT device across multiple retinal tissue samples provided by the research team
• Built Python preprocessing pipelines for raw OCT scan data including Gaussian denoising, CLAHE contrast enhancement, and image normalization to prepare clean 2D B-scan images for analysis
• Applied supervised ML models (Random Forest, SVM) on extracted image features from 2D B-scans to classify retinal conditions; improved ROC-AUC from ~0.65 to ~0.80–0.85 through iterative feature selection and class balancing
• model performance using ROC-AUC, sensitivity, and specificity metrics; plotted results using Matplotlib for research team review

B.E. - Electronics & Telecommunication - Savitribai Phule Pune University (2018)

Diploma - Electronics & Telecommunication - Govt. Polytechnic, Kolhapur (2014)