Content Discovery Workflow
Production content discovery workflow for an agentic recommendation platform, with optimized response time and measured search quality.
LangChainAzure AI SearchCosmos DB
15s → 6s response time
MM
MAHADEV MANERIKARAI ENGINEER PORTFOLIO
Available
AI Engineer | Backend-focused developer
Hi, I’mMahadev
Manerikar
I build production-grade AI systems, multi-agent platforms, and scalable backend APIs on Azure.
Pune, India · Open to opportunities
...Building since Mar 2024
2K+Documents ingested
60%Latency reduction · 15s to 6s
AzureProduction cloud
Continuum Control Plane
Multi-agent ServiceNow incident automation connecting support teams with Splunk signals and specialist agents.
LangGraphMCPWebSockets
Orchestrated incident workflows
DocVision-RAG
Multimodal retrieval for diagrams and screenshots, returning grounded answers with source pages and images.
Multimodal RAGVisionSSE
Searchable visual knowledge
Agent Orchestration Platform
Multi-user platform where database configurations become shareable, live agents and teams without redeployment.
AgnoFastAPISSE
Agents configured as data
// toolbox
Technical stack
The tools I use to design, build, evaluate, and operate production AI and backend systems.
AI & Agent Systems
LLMsAgentic & Multi-Agent SystemsLangGraphLangChainMCPMicrosoft Agent FrameworkAgnoAzure AI FoundryPrompt EngineeringGuardrails
LLMs
Agentic & Multi-Agent Systems
Microsoft Agent Framework
Agno
Azure AI Foundry
Prompt Engineering
GuardrailsRetrieval & Evaluation
RAGEmbeddings & Vector SearchAzure AI SearchPGVectorLLM-as-a-JudgeAzure AI Evaluation SDKRelevance & Groundedness
RAG
Embeddings & Vector Search
Azure AI Search
LLM-as-a-JudgeBackend & Data
PythonFastAPIREST APIsNode.jsReactWebSocketsSSESQLAlchemyPostgreSQLSQL ServerCosmos DB
WebSockets
SQL Server
Cosmos DBAzure & DevOps
Azure Container AppsAzure FunctionsLogic AppsApp ServiceAPIMKey VaultManaged IdentityMSALBlob StorageDockerAzure DevOps CI/CD
Azure Container Apps
Azure Functions
Logic Apps
App Service
APIM
Key Vault
Managed Identity
MSAL
Blob Storage
Azure DevOps CI/CDObservability
OpenTelemetryAzure Application InsightsAzure MonitorSplunkCost & Latency Tracing
Azure Application Insights
Azure MonitorLanguages
PythonJavaScriptTypeScriptSQL
// experience
Experience
Equations Work · Pune, India
AI and software engineering, from early prototypes to production systems.
AI Engineer
Equations Work · Pune
Build production multi-agent and RAG systems on Azure. Focus on orchestration, search quality, evaluation, observability, and performance—reducing a workflow response from 15 seconds to 6 seconds and search hallucinations from 8.2% to 2.2%.
- Moved the AVS Content Discovery Workflow from hosted agents to merged LangChain agents, reducing response time from 15 seconds to 6 seconds.
- Built agent evaluation and hallucination measurement; reduced the search tool hallucination rate from 8.2% to 2.2%.
- Designed specialist-agent orchestration for incident analysis, ServiceNow automation, and live chat streaming.
- Resolved production API timeouts at million-record scale with cursor-based pagination.
LangGraphAzure AI FoundryFastAPIMCPAzure AI Search
Trainee Software Engineer
Equations Work · Pune
Built Azure AI Foundry agents and Logic Apps integrations, and delivered proof-of-concept work for sales coaching and real-time vehicle tracking.
- Built three Azure AI Foundry agents for a sales coaching proof of concept.
- Connected Dynamics 365 and SQL Server workflows using Azure Logic Apps.
- Developed a real-time vehicle monitoring proof of concept with live tracking and PostgreSQL zone-duration logic.
PythonAzure AI FoundryLogic AppsPostgreSQL
Want more detail on the systems I’ve delivered at Equations Work?
// professional-projects
Equations Work projects
Selected AI and software projects delivered as part of my professional experience.
Content Discovery Workflow
Production agent workflow for personalized content recommendations.
- Moved from hosted agents to a merged LangChain architecture, reducing response time from 15 seconds to 6 seconds.
- Built agent evaluation and hallucination measurement, reducing the search tool hallucination rate from 8.2% to 2.2%.
- Added semantic search, language-based index selection, conversation state, and per-agent cost tracking.
LangChainAzure AI SearchCosmos DBApplication Insights
Incident Automation Platform
Coordinated agents that help support engineers investigate and resolve cloud incidents.
- Built specialist workflows for log analysis, incident similarity, root cause analysis, severity scoring, and ticket updates.
- Connected ServiceNow workflows through MCP services with streaming chat and operational observability.
LangGraphMCPFastAPIWebSocketsAzure
Enterprise Knowledge Assistant
Hybrid retrieval assistant for enterprise documents, with automated index ingestion.
- Combined vector and keyword retrieval across more than 1,000 documents to improve relevance and reduce hallucinations.
- Automated document ingestion and index updates with cloud functions and blob storage.
RAGAzure AI SearchAzure FunctionsBlob Storage
Property Intelligence Platform
AI-assisted property and investor research with automated reporting.
- Built data extraction, ETL, and ranking pipelines for financial and property insights.
- Automated investor-ready reports, saving more than eight hours of manual work per week.
Agentic AIETLDockerAzure DevOps
Sales Coaching Assistant
A proof of concept using specialized agents to support sales coaching workflows.
- Created orchestration, sentiment analysis, and email assistant agents.
- Connected Dynamics 365 and SQL Server data with workflow automation.
Azure AI FoundryLogic AppsDynamics 365
Real-Time Vehicle Tracking
Live vehicle monitoring dashboard with event-driven updates and notifications.
- Implemented live tracking, automated email notifications, and an event-refreshing dashboard.
- Added a PostgreSQL procedure to calculate time spent in each zone.
PostgreSQLReal-timeDashboard
// independent-work
Personal projects
Independent builds and experiments. Open a repository to explore the code.
agno-agents-hub
Agent Orchestration Platform
A multi-user platform where agent and team configurations live in a database and become runnable at request time, without redeploying.
- Users configure agent roles, instructions, tools, and model settings, then compose teams with a coordinator and ordered members.
- Runs execute outside the HTTP request and stream output over SSE; status, token usage, duration, and ownership are tracked.
AgnoFastAPISSEMulti-user
View repository DocVision-RAG
Multimodal document search
A RAG platform that makes PDF diagrams, flowcharts, and screenshots searchable alongside document text.
- Converts PDFs to Markdown, analyzes extracted images with a vision model, and places image analysis back in its original context.
- Uses header-aware chunking and cosine vector search to return cited pages and source images.
PythonVision modelsRAGAsync ingestion
View repository ServiceNow MCP Server
Agent-ready ticket workflows
An MCP server that gives AI assistants tools to retrieve and update ServiceNow incidents, with both stdio and SSE transports.
- Provides incident lookup and update tools through local stdio integration or an HTTP-based SSE server.
MCPServiceNowSSEPython
View repository Ingress AI
Unified LLM gateway
An OpenAI-compatible gateway that routes requests across LLM providers through one API, without changing client code.
- Routes models across providers with retries and automatic failover, plus caching and rate limits.
- Uses virtual keys and records token usage and costs in a shared ledger.
LLM GatewayOpenAI-compatible APIRoutingCost tracking
View repository // education-and-recognition
Education, awards & certifications
The foundations behind the systems I build today.
B.E. Computer Engineering
Zeal College of Engineering and Research, Pune · 2020–2024 · CGPA 8.90/10
Higher Secondary Certificate
SVM & Jr College of Science, Kudase, Sindhudurg · 2018–2020 · 77.23%
Unstoppable Contributor Award · 2025–26
Equations Work · Recognized for contributions to a production agentic AI project, including latency reduction, evaluation, and production defect resolution.
Certifications
Anthropic — Introduction to Model Context Protocol
DeepLearning.AI — LangChain for LLM Application Development (2024)
// contact
Let’s build something that ships.
Open to AI engineering and backend opportunities. Send a note or connect through one of the links.
Let’s build something amazing
Interested in collaborating on an AI project, have questions about my work, or want to discuss technology? Feel free to reach out.
LOCATIONPune, India
mahadev@portfolio:~$ help
Available: help, clear, /projects, /personal, /experience, /skills, /education, /contact, /home