NALLA SUMANG@nalla_sumang
2
Events attended
1
Submissions made
India
4+ years of experience
About me
An AI & Backend Systems Engineer specializing in the design and orchestration of multi-agent LLM workflows. The core engineering focus is on replacing single-threaded, blocking LLM calls with asynchronous, queue-based architectures that stay responsive under long-running inference. Expertise lies in system-level orchestration—using LangGraph and Google ADK to coordinate specialized AI agents, and structured outputs (Pydantic) to keep multi-step reasoning reliable rather than freeform. Rather than treating AI as a simple chatbot interface, the focus is on using AI as a reasoning layer embedded within decoupled services—agents that plan, retrieve, and act across separate frontend, backend, and worker processes. This technical execution includes deploying agent workflows over asynchronous message queues (Redis/BullMQ), maintaining persistent semantic memory through vector databases (Supabase pgvector, Pinecone), and containerizing backend and worker processes (Docker) for deployment across Render, Vercel, and Hugging Face Spaces
🤝 Top Collaborators
.png&w=640&q=75)
.png&w=640&q=75)
🤓 Latest Submissions
Sentinel: Autonomous AI Options Trading Agent
Retail traders and algorithmic developers face a massive challenge: digesting hundreds of breaking financial news headlines in real-time without emotional bias. Building automated pipelines that bridge unstructured news, LLM reasoning, strict capital risk gates, and broker API execution is traditionally complex and error-prone. Sentinel AI Trader solves this by introducing a 5-stage autonomous pipeline. First, it continuously ingests fresh market headlines and catalysts. Second, it leverages the blazing speed of Groq's Llama 3.3 70B model to analyze the sentiment and output a strict, structured JSON signal (determining Call/Put, Strike Price, and Confidence). The most critical component is the third stage: The Risk Gate. Before any order is placed, Sentinel audits the AI's signal against configured capital constraints, maximum portfolio concentration limits, and contract caps. If the AI's confidence is too low or the parameters are unsafe, the order is killed. If it passes, Sentinel generates the precise OCC Option Symbol and natively routes the market order exclusively to the Alpaca Paper Sandbox API. We built Sentinel with dual-stack parity: a Python data-science backend and a dynamic React/Vite web dashboard. To fulfill the hackathon's CLI requirement, we successfully embedded a fully interactive Alpaca CLI Terminal directly into our React UI, translating raw commands into real-time API requests. Sentinel demonstrates that bridging unstructured world events with strict, risk-gated execution APIs is the future of autonomous trading.
4 Sep 2026
