Surya Teja Reddy Dwarampudi

• RL environments for computer-use agents — high-fidelity GUI simulation so frontier models can learn to operate a desktop. Single-agent and multi-agent.
• SWE-Bench & Terminal-Bench tasks: authoring and curating evals that grade agents on real software-engineering work.
• Data pipelines turning messy real-world traces into training-grade data.
• Internal AI productivity tools: the stack the research team uses day-to-day.
• Partnering with 3+ global AI research labs to benchmark model performance. Mentoring junior engineers across workstreams.

• Real-time AI Voice Agent (IVR): low-latency calling platform on Gemini + Deepgram (ASR) + ElevenLabs (TTS) handling 15+ concurrent sales calls at <800ms latency.
• Sales AI Simulator: internal training platform where AI agents role-play as prospects, so reps can practice pitches before real calls.
• Automated Call Intelligence: compliance pipeline auditing 10K+ calls/month across sales and support. LLMs flag compliance issues and surface coaching insights.
• CRM Cost Optimization: architected in-house CRM replacing LeadSquared. Cut annual costs by 90% (₹5.4Cr+) with 100% adoption in 4 months.
• High-Scale Search: advanced filtering engine on 3M+ leads serving 100K+ queries/mo at sub-ms latency (ElasticSearch + tuned indexing).
• Zero-Conflict RBAC across 12+ modules · scalable monorepo (Turbo + Next.js) · no-code workflow builder on React Flow · double-digit concurrent A/B tests.

• Needle-Mover Award: reduced mentor no-shows from 11% → 3% via a custom scheduling feature on Ruby on Rails + React.
• Self-serve admin tools cut weekly support tickets by 60% (30→12), saving 100+ eng-hours/yr.
• Optimized the Article page with lazy-loading + server-side pagination, serving 500K+ MAU with improved core web vitals.

• Built interactive admin dashboards in Django to visualize inter-school activity data.

A low-latency calling platform that picks up, holds a real conversation, and books meetings. Built on Gemini + Deepgram + ElevenLabs, tuned for 15+ concurrent calls without sounding like an IVR.

Stack: Gemini, Deepgram, ElevenLabs, Node.js, Redis, WebRTC

Role: Lead engineer · architecture, latency, orchestration

• <800 ms — avg latency
• 15+ — concurrent calls
• 24/7 — uptime

Architected the in-house CRM that retired LeadSquared in 4 months. Zero-conflict RBAC across 12+ modules, sub-millisecond filtering on 3M+ leads, a no-code workflow builder, and double-digit concurrent A/B tests from day one.

Stack: Next.js, TurboRepo, ElasticSearch, PostgreSQL, React Flow, Kafka

Role: Frontend architecture · search infra · RBAC · workflow engine

• 90% — cost reduction
• 3M+ — leads indexed
• 100% — team adoption

Built the automated call-intelligence layer. Every sales and support call gets transcribed, classified, scored against a compliance policy, and fed back as a coaching nudge. LLMs catch the issues a manual QA pass tends to miss.

Stack: Python, FastAPI, OpenAI, AWS MSK, Postgres, OpenSearch

Role: Pipeline architecture · LLM prompting · review tooling

• 10K+ — calls/month
• 94% — precision flagging
• 3 d — review-to-prod loop

An internal training arena where AI agents role-play as buyers across personas, objections, and stalling tactics. Reps practice live, get scored, and walk into the real call already warmed up.

Stack: Gemini, TypeScript, Next.js, Postgres, pgvector

Role: End-to-end · prompts, eval harness, frontend

• 6 wks — rep ramp-up cut
• 40+ — persona library
• 5x — practice volume

RL environments (single + multi-agent) for computer-use agents, SWE-Bench and Terminal-Bench eval tasks, data pipelines, and the internal AI tools the research team runs on. Founding team — started Oct '25 inside Scaler, workspace spun out Apr '26.

Stack: Python, TypeScript, Playwright, Docker, RL · PPO/GRPO

Role: IC SDE2 · mentor to junior eng across workstreams

• 50+ — engineers in org
• 3 — research partners
• 7 — workstreams