Aalekh - A Solution Space Map Visualizer
Team of Stony Brook MS Data Science builders with experience at Deloitte, Oracle, and KPMG, specializing in RAG, agentic workflows, GNNs, and multimodal AI systems.
YouTube Video
Project Description
Aalekh — A Solution Space Map Visualizer
Aalekh is a spatial, timeline-driven reasoning experience that turns conversation into an explorable map of decisions. Instead of placing Claude inside a standard chat window, Aalekh makes the interface itself the innovation: users don’t just read answers—they manipulate Claude’s thinking as a living structure.
A session begins with a single problem statement. Aalekh then runs a focused interrogation loop across five constraint dimensions (resources, timeline, risk tolerance, market, founder context). Each user answer visibly “constructs” the solution space: constraint chips appear on the canvas, the map transitions through a fog-to-clarity system, and the experience culminates in an ignition moment where Claude generates a full node tree (12–15 nodes) with hierarchy, conflicts, explanations, and spatial positioning. This output isn’t a static visualization—Claude’s reasoning drives the schema that the UI renders in real time.
Prototype stability comes from a strict separation of concerns and defensive execution. Phase 1 (backend brain) uses LangGraph to model Claude’s work as discrete reasoning nodes (interrogator, map generator, expander, fork regenerator) operating over a canonical state object. Redis persists the session state and stores map snapshots per timeline index, enabling instant rewind and branch switching without recomputing. Every Claude response is schema-validated; malformed JSON triggers a controlled retry and graceful fallback.
Phase 2 (frontend experience) uses CopilotKit for real-time state sync (useCoAgent) and human-in-the-loop confirmation (renderAndWait). Aalekh shatters UI conventions through playful interaction: a timeline scrubber rewinds reasoning like video editing, and forking lets users branch alternate realities from earlier answers—making exploration the core loop, not chatting.
Technologies, Frameworks, Libraries:
Claude Sonnet via Anthropic API
LangGraph (Python) + LangChain
Redis (session persistence + snapshot versioning)
CopilotKit (useCoAgent, useCopilotAction, useCoAgentStateRender)
Next.js (frontend + runtime API routes)
React
SVG (edges/graph connections)
CSS transitions + keyframes (fog, ignition, panel choreography)
Team
Products & Tools
Additional Links
Github Repo with all our code
Drive link for the video