Hackathon work, research-oriented workflows and workshop prototypes. Each is labelled with what it is, not dressed up as production.
Hackathon project
Pricing Consequence Agent
Agentic AI · Decision support · Memory
An agent that analyzes pricing proposals using remembered patterns and previous outcomes, so future recommendations can be more informed by past decisions. My role: agent and backend engineer, with Laxmi (lead), Midhat and Neha.
- Agent response flow around a pricing proposal.
- Similar-experience recall and pattern reflection via a memory service abstraction.
- Structured outputs for responses, evidence, patterns and contradictions.
- APIs to ask, submit outcomes and fetch history; relevance thresholds, error handling, tests.
Python · FastAPI · Groq · Hindsight-style memory
No claim of better pricing decisions without measured evidence. Some integration and deployment work remains incomplete.
Experimental
arXiv Paper Digest & QA Agent
AI agents · Research automation · Retrieval
A research-oriented workflow that retrieves arXiv papers, ranks candidates, extracts content, summarizes it and supports question answering.
Query understanding → retrieval → ranking → parsing → chunk & embed → summary → Q&A
- Query broadening and retrieval retry.
- Candidate ranking and degraded-mode parsing.
- Chunking, embedding, and structured briefing and score models.
Groq API · Chroma · graph-based orchestration
Not evaluated for research accuracy; some features are still experimental.
Workshop prototype
Mastery Desk
Multi-agent AI experiment
A prototype from a multi-agent AI workshop: five specialized agents, two voice-oriented and three back-office, including a Google Calendar practice-session workflow and plan-related email generation.
An experiment, not a commercial product. Screenshots and links will be added once verified.