ojas@portfolio~/about
$whoami

Ojas Soni — AI Engineer

Role: AI Engineer

Building agentic AI systems. Wiring LLMs into production. Making retrieval actually retrieve.

About

$cat about.txt

AI Engineer working on agentic AI and retrieval systems that run in production. I build multi-agent pipelines, RAG platforms, and the FastAPI services that hold them together — currently automating incident resolution and candidate evaluation at enterprise scale.

Skills and tech stack

$ls skills/
  • PythonPython
  • LangGraph
  • LangChain
  • CrewAI
  • FastAPIFastAPI
  • PyTorchPyTorch
  • DockerDocker
  • KubernetesKubernetes
  • AWSAWS
  • MongoDBMongoDB
  • PostgreSQLPostgreSQL
  • DuckDBDuckDB
  • Kafka
  • GitGit
  • LinuxLinux
$cat links.txt

Experience

$cat experience.json | jq

Tata Consultancy Services

[ACTIVE]
AI EngineerNov 2024 → Present
Client: Apple
Noida, IN

AI-Driven Automated Ticket Resolution System

  • PythonPython
  • LangGraph
  • MCP
  • Kafka
  • FastAPIFastAPI
  • DockerDocker
  • - Engineered an agentic incident automation platform integrating MCP servers with a real-time Kafka ingestion pipeline processing 10,000+ production tickets daily, with RAG-powered orchestration agents that autonomously analyze and resolve production incidents.
  • - Developed specialized AI agents for Splunk log intelligence, anomaly detection, and contextual error summarization, integrating enterprise ITSM ticketing portals for real-time incident ingestion and tracking.
  • - Built FastAPI microservices with async endpoints for agent orchestration, health monitoring, and audit logging, containerized via Docker on scalable cloud infrastructure.
  • - Implemented a RAG-driven root cause analysis pipeline using historical incident embeddings, semantic retrieval, and structured remediation generation — cutting manual triage effort by 40% and average resolution time from hours to minutes.

AI-Powered Talent Evaluation and Interview Intelligence System

  • PythonPython
  • CrewAI
  • Gemini
  • FastAPIFastAPI
  • MongoDBMongoDB
  • - Built an agentic resume screening and interview automation platform on CrewAI and Gemini, with autonomous agents for resume parsing, dynamic question generation, and skill-gap analysis, backed by RAG retrieval of JD-specific questions from MongoDB.
  • - Implemented a multi-dimensional evaluation framework scoring candidates on correctness, clarity, depth, and relevance using Sentence Transformers for semantic similarity, reaching 87% agreement with human interviewer decisions.
  • - Built session management with FastAPI handling 100+ concurrent interviews, MongoDB persistence for conversation history, and sub-2s response latency across all agent interactions.
  • - Developed a summary generation agent producing structured hiring reports (strengths, skill gaps, risk factors, recommendations) with automated PDF generation via ReportLab, reducing end-to-end interview processing time by 60%.

Projects

$ls -la projects/

Hashnode Publishing Pipeline↗

  • PythonPython
  • GitHub ActionsGitHub Actions
  • GraphQLGraphQL
  • - Built a Git-based publishing workflow that turns a plain repository into a Hashnode CMS — writing a post is a git push, with no dashboard in the loop.
  • - A GitHub Actions job diffs each push for changed markdown under posts/, parses YAML frontmatter for title and tags, and publishes through Hashnode's GraphQL API by creating a draft and then publishing it.
  • - Persisted post IDs in a mapping file so re-pushing an edited file updates the live post in place instead of publishing a duplicate.

Enterprise Secure Multi-Tenant RAG Assistant

  • PythonPython
  • LangChain
  • FastAPIFastAPI
  • DockerDocker
  • - Architected a secure multi-tenant Agentic RAG platform with RBAC-driven document-level authorization, metadata-aware retrieval, and enterprise-grade knowledge isolation across dynamic document repositories.
  • - Engineered scalable LLMOps pipelines for automated document ingestion, chunking, embedding orchestration, and hybrid semantic retrieval leveraging vector databases, reranking, and real-time synchronization workflows.
  • - Integrated open-source LLMs with citation-grounded generation, conversational memory, audit telemetry, and high-throughput FastAPI-based inference orchestration for production-scale enterprise AI assistants.

Education

$cat education.txt

Thapar Institute of Engineering and Technology

B.E. Computer EngineeringAug 2020 → July 2024
Punjab, IN
$