Aakash Mahesha

About

Built and productionized a tool-first, multi-agent framework that augments an anomaly detection model with domain context to generate trustworthy, evidence-backed anomaly explanations (including false-positive likelihood). Architected the platform to be model/orchestration/vectorDB agnostic (e.g., GPT + CrewAI + ChromaDB vs Claude + LangGraph + other vector DB) with strong performance, reliability, and OpenTelemetry-based observability. Also built a personal LangGraph-based "mock interviewer" agent that asynchronously fuses voice + live code input using state reducers, stop conditions, and fallback routing.

Experience

Education

Key Strengths

• Designed and deployed model/orchestration/vectorDB-agnostic multi-agent service using Factory Method abstractions
• Performance optimization for agent systems (O(1) initialization, connection pooling, multi-layer caching, async/concurrent tool pipelines)
• Reliability engineering for LLM/tooling dependencies (timeouts, retries/backoff, circuit breakers, fallbacks)
• Strong observability implementation (structured logs, metrics, OpenTelemetry tracing; golden signals + domain metrics)
• Stateful workflow orchestration with LangGraph (parallel streams, reducers, stop conditions, gating rules, fault isolation)
• Production-minded agent design with clear contracts, guardrails, and measurable evals (success rate, groundedness, latency, cost)
• Cross-functional collaboration translating non-technical triage needs into acceptance criteria and structured, evidence-grounded explanations