Sayan Pramanick

Problem: Professionals were losing hours to email backlog, scattered meeting notes, and manual follow-ups. The MVP had to automate inbox triage, reply drafting, and meeting summarization while preserving user trust with review-first workflows, secure integrations, and no autonomous outbound actions.

Scalability solved: Structured the platform as a modular assistant system so the same core services could support single-inbox users, multi-account consultants, and enterprise teams with SSO, RBAC, and analytics.

Latency optimization: Background workers handled classification and draft generation within seconds of email arrival, while the meeting pipeline produced summaries and follow-up drafts within minutes after a call.

Cost strategy: The system stayed draft-only and suggestion-led, keeping users in control while corrections improved personalization and reduced administrative workload without risky full automation.

Problem: Ecogensus needed a fast, explainable way to handle highly variable waste composition while still meeting target fuel specs. I had to ship a functional MVP in 30 days, support auditable recommendations, and avoid slow manual trial-and-error across process variables.

Scalability solved: Architected as a scalable prototype for multi-facility rollout with reusable APIs and modular model components.

Latency optimization: Enabled real-time scenario recalculation in the dashboard so operators can test composition and process changes instantly.

Cost strategy: Reduced costly trial runs and manual analysis cycles by surfacing prediction and optimization guidance before plant-level execution.

Problem: Auxilo needed to replace fragmented, paper-heavy workflows with a single digital engine. Each application required reconciling multi-party profiles, processing heterogeneous documents (identity, income, collateral, academic, visa), and passing strict KYC, AML, and bureau checks, which introduced delays and manual keying risk.

Scalability solved: Deployed containerized services on AWS (EC2 + ECR) with Redis-backed async workers and persistence split across PostgreSQL and MongoDB to handle parallel loan files and partner traffic spikes.

Latency optimization: API-driven checks and straight-through extraction shifted verification from batch/manual turnaround to near real-time decision preparation within the same session.

Cost strategy: Automating extraction, validation, and compliance orchestration removed repetitive back-office keying and review loops, lowering per-file processing cost while improving auditability.