MKManish
Email me

Manish Kumar Prajapati

Senior Software Engineer · Applied AI Engineer · Java, AWS & Agentic AI

const manish = 12y Java · Agentic AI · RAG · AWS certified · MSc AI/ML @ LJMU

12 years building Java and Spring Boot systems on AWS. Now I build AI into them: agents that call tools, RAG over company documents, and LLM chatbots in production. I have an Executive PG in ML and AI from IIIT Bangalore and complete my MSc in AI and ML at LJMU in 2026. I also build and run my own AI products.

View projects Contact me
manish — zsh
$ java -version
java 25
$ ls products/
coder/livecoder.lyzentra.com ↗
emerjo/liveemerjo.lyzentra.com ↗
srclens/livesrclens.lyzentra.com ↗
pulse/livepulse.lyzentra.com ↗
ombriq/in progress
$ cat stack
Java 25 · Spring AI · LangGraph · Bedrock
$ cat certs
aws solutions architect · aws ai practitioner
$
scroll

01 / projects

Products I build

Five Lyzentra products, from code to the server they run on. Four are live and Ombriq is on the way.

lyzentra.com
ai agentLive

Lyzentra Coder

Ticket to pull request

An agent that reads a ticket, writes the change in the right repository, builds and tests it, and opens a pull request for a person to review. Runs on a free local model.

local model · build + test check · human reviews every PR

  • AI agent
  • LLM
  • Tool calling
  • MCP
  • Local LLM
  • Java 25
  • Spring Boot 4
  • Angular
  • Ollama
  • Docker
trackerLive

Emerjo

Open AI models rising this week

Tracks new open models across Hugging Face, GitHub and arXiv, and ranks what is gaining ground.

Hugging Face · GitHub · arXiv · weekly ranking

  • AI agent
  • LLM chat
  • Streaming answers
  • Multi-model fallback
  • Model ranking
  • Spring Boot
  • PostgreSQL
  • Angular
dev toolLive

SrcLens

A repository as diagrams

Turns a code repository into diagrams of its parts, calls and request flows. Every arrow links to the file and line that proves it.

parts, calls, request flows · every arrow cites file:line

  • AI agent
  • Local LLM
  • AI naming and grouping
  • AI chat over code
  • Java
  • Parsers
  • Angular
monitoringLive

Pulse

AI model status, measured from India

Checks AI model APIs from Mumbai on a schedule: latency, failures and a daily quality score.

probes from Mumbai · latency + failures · daily quality score

  • AI agent
  • LLM benchmarking
  • Latency and failure tracking
  • Answer quality scoring
  • Java 25
  • Spring Boot
  • Scheduling
videoIn progress

Ombriq

Product promo videos

Give it a product link and a few notes, get a vertical reel with script, voiceover and captions.

script · voiceover · captions · vertical reel

  • AI agent
  • LLM script writing
  • Speech-to-text
  • AI voiceover
  • Local LLM
  • Java 25
  • Spring AI
  • Ollama
not public yet

02 / reels

See them in action

Short product reels, made with Ombriq, my own AI video tool.

Music: Undaunted by Kevin MacLeod (incompetech.com), licensed under CC BY 4.0.

03 / ml work

ML and AI work on GitHub

Projects from my Executive PG in ML and AI at IIIT Bangalore. Code and notebooks are public.

rag · llmgithub.com/manishkprj/policy-insurance-rag

Insurance policy Q&A (RAG)

problemAgents and customers lose time reading long policy PDFs, and keyword search misses the answer.

  • Upload policy PDFs, split them into nodes with sentence or semantic splitters, and embed them into ChromaDB.
  • A router query engine picks the right retrieval pipeline; results are reranked before the answer is generated.
  • Every answer cites the source document and page.
  • Switch the embedding model, chunk size and context window live from the UI, with no restart.
  • Disk cache for repeated questions and a response evaluation module.

chatbot · llmgithub.com/manishkprj/shopassist-ai

ShopAssist AI: laptop advisor chatbot

problemPicking a laptop means comparing specs most buyers do not understand.

  • The chat draws out six needs: GPU, display, portability, multitasking, speed and budget.
  • GPT-4o function calling turns the conversation into structured preferences.
  • GPT-4o classifies every laptop's raw specs into the same features ahead of time.
  • Scoring filters by budget, adds a point per matching feature, keeps scores above 2 and returns the top 3.
  • Every message passes the OpenAI Moderation API; API calls retry with exponential backoff.

nlp · recommendergithub.com/manishkprj/sentiment-based-product-recommendation-system

Sentiment-based product recommendations

problemAn online store wants recommendations customers will actually like, not just similar items.

  • Cleaned and lemmatised reviews; TF-IDF with 1–3-word n-grams and 10,000 features.
  • Compared logistic regression, random forest, naive Bayes and XGBoost with SMOTE and grid search; logistic regression gave the best balance.
  • Item-to-item collaborative filtering beat user-to-user on RMSE and coverage.
  • Takes each user's top 20 products and keeps the 5 with the most positive review sentiment.
  • Served through a Flask web app with username autocomplete.

deep learning · cnngithub.com/manishkprj/MelanomaSkinCancerDetection

Melanoma detection with a CNN

problemMelanoma causes most skin cancer deaths; automatic screening of skin images can save dermatologists time.

  • Custom convolutional network in TensorFlow for 9 lesion classes from the ISIC dataset.
  • Five experiments: baseline, data augmentation, dropout, batch normalisation and class rebalancing.
  • Class rebalancing fixed the overfitting and lifted validation accuracy from 55% to about 86%.

regressiongithub.com/manishkprj/BikeSharingLinearRegression

Bike-sharing demand model

problemA bike-sharing company needs to know what drives daily demand, to plan for it and for new markets.

  • Explored season, weather, month, holiday and weekday effects on rentals.
  • Dropped collinear and leaky columns and encoded the categorical ones.
  • Built a multiple linear regression and pruned features step by step.
  • Temperature and year turned out to be the strongest positive drivers of demand.

eda · riskgithub.com/manishkprj/LendingClubCaseStudy

Lending Club loan default analysis

problemA lender loses money when risky borrowers default. Which factors predict it?

  • Cleaned the Lending Club loan data and analysed it: univariate, segmented, bivariate and multivariate.
  • Higher interest rates, grades E to G, 60-month terms, high debt-to-income and low income go with more defaults.
  • Turned the findings into combined risk rules, such as high debt-to-income plus a large loan on a 60-month term.
View all on GitHub

04 / stack

What I work with

AI first: agents, LLMs and ML, built on 12 years of Java and AWS.

ai · primary focus

AI agents, LLM apps and machine learning, backed by an Executive PG from IIIT Bangalore, an MSc at LJMU (2026) and AWS AI Practitioner.

  • AI agents
  • Multi-agent systems
  • RAG
  • LLM apps
  • MCP
  • Tool calling
  • LangGraph
  • Spring AI
  • AWS Bedrock
  • Prompt engineering
  • LLM evaluation
  • LLM security
  • NLP
  • Deep learning
agents/9 tools

AI agents

Agents that plan, call tools and hand work to a person for review. Lyzentra Coder is one.

  • Agent design: tool calling, planning, memory
  • Multi-agent orchestration: LangGraph, AutoGen, CrewAI
  • MCP servers and tools
  • Agents for Amazon Bedrock
  • Spring AI tool calling
  • Coding agents: ticket to pull request
  • Human-in-the-loop review
  • Guardrails and safety gates
  • Agent tracing and evaluation
llm/12 tools

LLMs and RAG

LLM features that run in production.

  • AWS Bedrock (Claude, Titan, Knowledge Bases, Agents)
  • OpenAI (GPT-4o, embeddings, function calling)
  • Spring AI
  • RAG: hybrid search, re-ranking, query expansion
  • Semantic Kernel
  • LangChain
  • LlamaIndex
  • Hugging Face
  • Vector stores: ChromaDB
  • LLM evaluation: RAGAS
  • Prompt engineering
  • Ollama and local models
ml/10 tools

Machine learning

MSc in AI and ML (LJMU, 2026) and Executive PG in ML and AI (IIIT Bangalore).

  • Supervised and unsupervised learning
  • Deep learning: CNNs
  • NLP: NER, POS tagging, classification, sentiment
  • Text embeddings and semantic search
  • scikit-learn
  • TensorFlow
  • Hugging Face Transformers
  • EDA and feature engineering
  • Model training, cross-validation, evaluation
  • LLM security: malicious prompt and adversarial attack detection (MSc thesis)

foundation · java, cloud, data

backend/13 tools

Backend and system design

Where most of my years went.

  • Java 8 to 25
  • Spring Boot
  • Spring Security
  • Microservices
  • Clean architecture
  • Domain-driven design
  • REST API design
  • Event-driven: Kafka, SNS/SQS, EventBridge
  • JPA and Hibernate
  • MyBatis
  • Python and FastAPI
  • OpenTelemetry and distributed tracing
  • Resilience: Resilience4j, circuit breakers
cloud/10 tools

Cloud and DevOps

AWS certified. I run my own servers too.

  • AWS: EC2, S3, Lambda, ECS/EKS, RDS, DynamoDB, IAM
  • API Gateway
  • CloudWatch
  • Secrets Manager
  • Docker
  • Kubernetes
  • GitHub Actions (CI/CD)
  • Structured logging and observability
  • SonarQube quality gates
  • Linux, Caddy, Cloudflare
data/6 tools

Data

Relational, document and vector.

  • PostgreSQL
  • MySQL
  • Oracle
  • SQL Server
  • MongoDB
  • DynamoDB
frontend/8 tools

Frontend and testing

Full stack when the job needs it.

  • Angular and RxJS
  • React
  • TypeScript
  • JavaScript
  • JUnit
  • Spring Test
  • Cypress
  • Playwright

05 / services

What I can build for you

For a team that is hiring, or a project that needs doing.

AI agents and automation

Agents that call your tools and APIs, plan multi-step work and hand the result to a person to approve.

  • LangGraph
  • Spring AI
  • MCP
  • Bedrock Agents

RAG and LLM features

Search and Q&A over your own documents with sources cited, and chatbots for web, WhatsApp and more.

  • LlamaIndex
  • LangChain
  • ChromaDB
  • OpenAI

Java and Spring Boot backends

Microservices, REST APIs and event-driven systems with Kafka, written to be tested and kept.

  • Java 21+
  • Spring Boot
  • Kafka
  • MongoDB

AWS cloud architecture

Design and set up services on ECS, EKS, Lambda and API Gateway, as a certified Solutions Architect.

  • ECS/EKS
  • Lambda
  • API Gateway
  • Docker

Full-stack product builds

From idea to a live product: front end, API, database and hosting. I run five products of my own.

  • Angular
  • React
  • PostgreSQL
  • CI/CD

Modernising older systems

Bring legacy apps up to date: Java 8 to 21+, AngularJS to Angular, and reverse engineering what has no docs.

  • Java upgrades
  • AngularJS to Angular
  • Refactoring

06 / education

Degrees and certifications

Formal study, and the cloud exams I have passed.

  • certified

    AWS Certified Solutions Architect

    Amazon Web Services

  • certified

    AWS Certified AI Practitioner

    Amazon Web Services

  1. MSc, Artificial Intelligence and Machine Learning

    Liverpool John Moores University, UK

    Period
    Expected 2026

    Machine learning, NLP and applied AI, with research in LLM security.

    thesisContextual Analysis of Malicious Prompts for Detecting Adversarial Attacks on LLMs

    • LLM security
    • Adversarial attack detection
    • Prompt injection and jailbreaks
    • Transformers and LLMs
    • Machine learning
    • Deep learning
    • NLP
    • Model evaluation
    • Research methods
    • Python

  2. Executive PG, Machine Learning and AI

    IIIT Bangalore

    Period
    2025

    CGPA 3.83 / 4.

    • Machine learning
    • Generative AI and LLMs
    • RAG with LlamaIndex
    • OpenAI function calling
    • Prompt engineering
    • Deep learning: CNNs
    • NLP: sentiment, NER, POS tagging
    • Recommender systems
    • Supervised and unsupervised learning
    • Linear and logistic regression
    • Random forest, naive Bayes, XGBoost
    • Class imbalance: SMOTE
    • Model evaluation and cross-validation
    • Hyperparameter tuning
    • Statistics and EDA
    • Feature engineering
    • scikit-learn, TensorFlow, Pandas

  3. B.E., Computer Science

    IPS Academy, Indore

    Period
    2014

    Final-year project, a job portal in Java, JSP and MySQL, took first place at the college project contest.

07 / recognition

Awards and recognition

From the teams I have worked with.

  • Performance Star Award

    For results on the chatbot platform.

    2021 · 2023 · 2025

  • Long Service Award

    Five years of steady delivery and teamwork.

    2025

  • Performance Excellence Award

    E-commerce platform work.

    2019

  • Performance Excellence Award

    Emissions evaluation application.

    2017

  • First place, UDAAN project contest

    Job portal in Java, JSP and MySQL, IPS Academy.

    2014

08 / contact

Say hello

Hiring for a Java role, or need something built? Email is quickest.

[email protected]LinkedInGitHub