Open to AI engineering roles
Lakshya Karira
Agentic systems, multi-agent orchestration, and RAG that survives evaluation.
Index
Page 2 — About
AI engineer specializing in agentic systems — multi-agent platforms that plan, coordinate, and execute tasks with minimal supervision. Experience spans two AI engineering internships, at Capmob Financial Services and Stremly, building and shipping these systems in production rather than as prototypes.
Particular focus on two areas production agent systems often get wrong: scoping what an agent is actually allowed to do, and evaluating retrieval and reasoning pipelines rigorously before they ship. Comfortable owning a system end-to-end, from architecture through the tooling that keeps it reliable once live.
Page 3 — Experience
Capmob Financial Services
Jun 2026 – Sep 2026AI Engineer Intern · Bengaluru, India
A two-sided agentic lending platform — ten agents, one matching engine between them.
- Helped architect a 10-agent, two-sided Agentic AI lending platform end-to-end: 5 borrower-side agents (document intelligence, credit narrative, deal tracking, education, CapBot) and 5 lender-side agents (deal discovery, risk intelligence, portfolio, compliance, preference engine), connected via a central matching engine.
- Owned CapBot's document intelligence pipeline end-to-end — RAG-based classification and extraction across bank statements, trade licenses and salary certificates, with structured dataframe output and PII guardrails ahead of downstream credit analytics.
- Managed the platform's AWS Bedrock integration (Nova Micro) powering the agent pipeline's LLM calls, including CloudWatch monitoring, alarms and log tailing for production reliability.
RAG · AWS Bedrock · Nova Micro · CloudWatch · Multi-Agent · PII Guardrails
Stremly
Apr 2026 – Jun 2026AI Engineer Intern · Bengaluru, India (Remote)
Six agents planning and executing projects on their own — plus the tooling to prove they actually worked.
- Architected and validated a 6-agent multi-agent system for end-to-end project execution, enabling autonomous planning, reasoning and task orchestration across production workflows.
- Built an API status monitoring page for agentic pipelines, reducing time-to-detect failed production APIs from hours to minutes and enabling faster incident response.
- Designed API integration test suites covering inter-agent communication and external services, improving the reliability of production agent workflows.
Multi-Agent · LangGraph · Monitoring · Integration Testing · Python
Page 4 — Projects
Onboarding Assistant
2026A benchmarked multi-agent RAG system
Three RAG architectures, built and measured against each other instead of one built and hoped for.
LangGraph · FAISS · BM25 · Groq · RAG-Eval ↗
Custos Gate
2026A trust and authorization layer for agent-to-agent commerce
Razorpay lets an agent move money. This decides whether it was allowed to — and proves it afterwards.
TypeScript · Razorpay MCP · Razorpay Route · Policy engine · Agentic-Payment ↗
CyberCortex
2026Autonomous SOC analyst
An agentic loop that reconstructs multi-stage attacks across on-premise and AWS, and keeps going until it is confident.
LangGraph · Elasticsearch · AWS CloudTrail · MITRE ATT&CK · CyberCortex ↗
Self-Healing Sandbox
2026Autonomous QA agent
Paste a bug report, get a reproducible Dockerfile — with a test that repairs its own selectors when the page moves.
Python 3.11 · Gemini 2.5 · E2B · Playwright · Self-Healing-SandBox ↗
Compiler Error Explainer
2026A decoder-only transformer, from scratch
A 6.9M-parameter GPT-style language model built without pretrained weights, trained to explain compiler errors in English.
PyTorch · Custom BPE · HuggingFace Tokenizers · Kaggle T4 · Compile-Error ↗
Page 5 — Achievements
2nd Place — AI-SEC Community Hackathon
Sep 2026AI-SEC Community
Built AgentID, a trust layer for AI agents that revoked an agent's access in under a second on stage.
DID · Verifiable Credentials · Agent Identity · Zero Trust
3rd Place — Inter-College Hackathon
Dec 2024DSCE
Built an AI Driver Assistance System using YOLO and OpenCV for real-time dizziness and obstacle detection.
YOLO · OpenCV · Real-time Detection
Page 6 — Skills & Education
Agentic systems
LangGraph · LangChain · Model Context Protocol (MCP) · Multi-agent orchestration · Retrieval-Augmented Generation · Hybrid search (FAISS + BM25) · LangSmith
Databases
MongoDB · PostgreSQL · Redis · ChromaDB · Elasticsearch
Cloud & infrastructure
AWS Bedrock · AWS ElastiCache · AWS CloudTrail · Docker · GitHub Actions · CI/CD · REST APIs · OAuth
Machine learning & NLP
PyTorch · Transformers · HuggingFace · scikit-learn · NumPy · Pandas
Languages & tools
Python · TypeScript · Java · Git · Jupyter · Postman
Dayananda Sagar College of Engineering
B.E. Computer Science · Bengaluru, India · Expected Aug 2028
Page 7 — Contact
Open to AI engineering roles · Bengaluru, India
— that's the whole notebook.