Tech Stack
The Problem
Navigating modern AI and machine learning literature has become overwhelming due to the exponential publication velocity on ArXiv. Traditional keyword searches return disconnected paper lists without surfacing underlying citation dynamics, conceptual lineage, or contradictory findings across research groups. The challenge was to build an autonomous agent that systematically parses, filters, and maps semantic relationships between papers at machine speed.
Architecture
ThesisLens runs an autonomous research intelligence pipeline orchestrated by a 6-step LangGraph state machine powered by Groq's Llama 3.1 8B Instant model for sub-second structured inference. The workflow dynamically plans search strategies across ArXiv, fetches matching candidate papers, filters them for domain relevance, detects inter-paper relationships, analyzes unaddressed research gaps, and synthesizes a structured field report. Real-time execution telemetry is streamed from a FastAPI backend over Server-Sent Events directly to the client interface. The React frontend visualizes the resulting literature topology using an interactive D3.js force-directed knowledge graph, where node size reflects connection count, node color encodes research category, and typed edges map semantic relationships such as extends, contradicts, related, and replicates.
Approach
- 01
Orchestrated a 6-step LangGraph state machine from search strategy planning through multi-hop ArXiv fetching, relevance filtering, and field report generation.
- 02
Integrated Groq (Llama 3.1 8B Instant) for sub-second LLM inference, enabling rapid multi-paper relationship classification and synthesis.
- 03
Implemented FastAPI Server-Sent Events (SSE) streaming to transmit stage-by-stage pipeline telemetry to the frontend in real time.
- 04
Built an interactive D3.js force-directed knowledge graph mapping paper connectivity, domain categorization, and semantic link types.
- 05
Developed an automated research gap detection module that highlights unresolved questions across related papers.
Outcome
An autonomous research intelligence platform that compresses hours of literature discovery into seconds. By pairing low-latency Groq inference with dynamic D3.js knowledge graph visualization, ThesisLens enables researchers to explore complex multi-paper relationships and identify novel research gaps with full citation provenance.