Negotia – Agentic AI Negotiation Coach with Adaptive Memory
Agentic AI negotiation coaching platform with scenario generation, adaptive AI opponents, real-time conversations, structured feedback, cross-session memory, and personalized strategies. Built with LangGraph, FastAPI, React, PostgreSQL, and Amazon Be
Business Challenge
Negotiation is a critical professional skill, yet opportunities to practice are limited. Real negotiations involving compensation, promotions, clients, or vendors carry financial and professional consequences. Traditional role-playing requires another participant, professional coaching can be expensive, and static simulations lack realism and personalized feedback. The challenge was to build an accessible AI-powered practice environment that could simulate realistic negotiations, maintain conversation state, evaluate user behavior, deliver actionable coaching, and adapt across multiple sessions. The system also needed reliable orchestration, persistent user data, secure access, and a deployable full-stack architecture.
Solution Delivered
Designed and deployed Negotia, a full-stack agentic AI negotiation coaching platform using Python, FastAPI, React/TypeScript, PostgreSQL, LangGraph, and Amazon Bedrock. Implemented six specialized AI components: Scenario Generator, Opponent, Coach, Debrief, Strategy, and Negotiator Memory. A deterministic negotiation engine coordinates live interactions, while LangGraph orchestrates the session-completion workflow. Built streaming opponent responses, structured negotiation-state extraction, silent coaching observations, personalized debriefs, actionable strategies, and cross-session memory that adapts future coaching and opponent context. Implemented Pydantic-validated outputs, JWT authentication, user-scoped data access, idempotent completion, transactional persistence, and automated tests. Containerized and deployed the application on AWS EC2 using Docker Compose and Nginx, with FastAPI and PostgreSQL isolated within the application network. GitHub: https://github.com/bikash-singhal/negotia
Outcomes Achieved
Delivered a working, deployed AI negotiation coaching application that supports the complete practice lifecycle: scenario creation, interactive negotiation, behavioral analysis, personalized debrief, strategy generation, and cross-session coaching memory. Implemented six specialized AI reasoning components coordinated through deterministic application logic and a LangGraph completion workflow. Improved workflow reliability through strict structured-output validation, safe streaming persistence, authenticated resource ownership, idempotent completion requests, and atomic persistence of coaching artifacts. Enabled personalized learning across sessions through a bounded coaching profile that tracks recurring strengths, weaknesses, improving skills, persistent risks, and recommended next-session exercises. Established a testable deployment architecture using a deterministic fake LLM provider, automated tests, PostgreSQL persistence, Docker, and AWS EC2.
Measurable Business Outcome
Implemented and deployed an end-to-end AI coaching workflow comprising six specialized AI components: Scenario Generator, Opponent, Coach, Debrief, Strategy, and Memory. The platform delivers five core stages of negotiation practice: scenario generation, interactive simulation, personalized debrief, actionable strategy, and cross-session coaching. Reliability improvements include idempotent session completion, atomic persistence of coaching artifacts, strict structured-output validation, and safe handling of interrupted streaming responses. Cross-session memory maintains a structured coaching profile containing stable strengths, weaknesses, improving skills, persistent risks, and a prioritized next-session exercise. These are verified implementation outcomes; customer adoption, learning improvement percentages, and financial impact have not yet been measured.