Sanjit Jeevanand

Who it's for Buy-side and sell-side analysts, equity researchers, and compliance teams querying UK-listed company filings and regulatory disclosures.

Problem Analysts spend 40%+ of time searching and synthesizing scattered financial documents. Critical insights buried in 100+ page regulatory filings are missed or delayed.

Decisions it supports Risk analysis, document interpretation, regulatory insight extraction, and compliance verification. Reduces research time from hours to minutes with grounded, cited responses.

System Overview End-to-end RAG system over UK annual reports and regulatory filings. PDFs are ingested, page-aware chunked, and embedded using SentenceTransformers. FAISS enables sub-second semantic retrieval. A grounded generation layer assembles evidence contexts and produces citation-backed answers. Deployed as a stateless FastAPI service on Google Cloud Run with autoscaling.

Tech Stack

• Python
• SentenceTransformers
• FAISS
• FastAPI
• OpenAI API
• GCP Cloud Run + Docker

Who it's for Applied ML teams operating predictive models in production who require reliability, monitoring, and governed model lifecycle management.

Problem Model degradation in production often goes undetected for weeks. Teams lack systematic signals for retraining decisions, leading to stale models or unnecessary compute spend.

Decisions it supports Retrain vs. hold, deploy vs. rollback, resource allocation. Automated alerting on data drift, concept drift, and performance decay with statistical significance testing.

System Overview Production ML platform covering offline training, artifact-based model versioning, online inference, and continuous monitoring. FastAPI serves low-latency predictions while asynchronous jobs compute data and prediction drift using KS and PSI metrics. Threshold breaches trigger retraining and controlled model promotion workflows on AWS.

Tech Stack

• Python
• FastAPI
• XGBoost / LightGBM
• Docker
• AWS (ECS / EventBridge)
• Statistical Drift Metrics (KS, PSI)

Who it's for Engineering and ML teams evaluating high-impact architectural and system design decisions under competing constraints.

Problem Complex architecture decisions involve competing constraints (latency, cost, maintainability) that are rarely quantified systematically. Teams default to highest-paid-person's opinion or analysis paralysis.

Decisions it supports Architecture choices, infrastructure tradeoffs, ML system design patterns. Provides structured evaluation frameworks, cost projections, and audit trails for decision rationale.

System Overview Agentic decision system composed of planner, specialist, critic, and synthesizer agents coordinated via a deterministic state machine. Agents execute concurrently, surface disagreements, and produce an auditable decision trace with confidence scores. Real-time interaction is provided via WebSockets, with explicit token-level cost accounting and autoscaling deployment on Cloud Run.

Tech Stack

• Python (asyncio)
• FastAPI + WebSockets
• LangGraph + LangChain
• OpenAI API
• Docker + Google Cloud Run
• Token-level cost tracking