11 years architecting ML and AI systems at scale — from 0-to-1 products to production serving millions. I help early-stage teams design, build, and ship AI that actually works.
The problem: commercial-real-estate account managers juggling brokers, deals, documents, and email/SMS scattered across inboxes.
A greenfield, multi-tenant platform on AWS — FastAPI on ECS Fargate, Aurora Serverless v2, Cognito — with a unified email/SMS inbox (Missive + Twilio) and a self-healing bug pipeline that dedupes with pgvector and opens a fixed PR via a bounded Claude Code session.
The problem: a Talent Operating System — the “mind of the company” — that has to learn as people evolve across assessment cycles.
An event-driven architecture — transactional outbox → SNS/SQS FIFO, ordered per person — over a Neptune temporal knowledge graph, with scope-enforced Playbook and Coach APIs across Individual, Manager, and HR views.
The problem: wealth, health, and longevity live in separate silos — no single view of whether you're actually on track.
A self-built platform ingesting accounts via Plaid and biometrics from Oura, then running a longevity survival model, Monte Carlo retirement simulation, and a rule-based risk engine into one retirement-readiness score. See the live demo →
The problem: generating enough high-quality experiment variants — and shipping them — was bottlenecked on hand-coding.
An LLM pipeline that segments a page and ideates 10+ validated variants per site, paired with an automated code-gen framework — WebSocket → Lambda → Claude → production code — that cut deploys from hours to 10 minutes.
The problem: slow loan approvals at bank scale, under hard regulatory and interpretability constraints.
Predictive adjudication models — Random Forest features into CatBoost — running at 1M+ applications a month. They grew the loan book by $1B CAD and revenue by $40M CAD.
Own the technical vision, architecture, and delivery — from first commit to a deployed, maintainable platform.
Set the AI strategy and system design — data models, agents, evaluation — so what you build holds up in production.
Hands-on build of agentic and ML products end to end — greenfield to shipped, on AWS and modern LLM stacks.
Architecture reviews, hiring, and technical direction for teams who need a senior sounding board on AI.
I'm Thejas Prasad — an engineer with a contemplative mind. Eleven years building ML and AI, from a founding-team data scientist to fractional technical leadership across startups and enterprise.
Manas is Sanskrit for “mind.” I care about systems that are calm, legible, and durable — the same attention I bring to a Vipassana sit or to hosting a room full of people. I take on a small number of teams at a time so the work stays deep.
I take on a few teams at a time. Tell me what you're building and where it's stuck — a good first call is 30 focused minutes.