Verifiable output. Deterministic grading. Traceable failure modes.
Currently: Taking pilot engagements: compliance audits, eval-harness builds, and LLM QA pipeline architecture for collections, direct sales, and BPO verticals.
-
auditguard-mcp — A compliance-aware MCP server. Seven-step pipeline: RBAC, PII detection, policy enforcement, audit logging. 15-case eval, 100% pass. Live demo
-
Scrutiny — FDCPA/Reg F call transcript audit in 60 seconds. 12-rule rubric with verbatim evidence quotes and statutory citations. Dual-path evaluator. Live demo · Blog post
-
RegTriage-OpenEnv — An OpenEnv RL environment where the reward signal is auditor approval. 12 tasks, severity-weighted F1, auto-fail caps. Live demo · Blog post
-
LLM Deploy Cost Calculator — Production-grade GPU sizing, cost comparison, and break-even analysis for LLM deployment. Architecture-aware VRAM (GQA, MLA, MoE), throughput model, replica multiplier, pricing tiers. 49 model variants, 38 API plans. Live tool · Blog post
-
Inference Bench — Reproducible vLLM vs SGLang vs llama.cpp benchmark on NVIDIA L4 (via Modal). Concurrent-request sweeps, TTFT/TPOT, tail latency (p95/p99), success rate. SGLang leads throughput (+10%), vLLM leads TTFT at low concurrency. Showcase
Previously shipped LLM features to enterprise production at a speech analytics company serving regulated contact centers: automated quality scoring, real-time compliance assistant, conversational analytics engine. Self-hosted on customer infrastructure. Zero data egress.
rituraj.info: notes on production ML, compliance systems, and agent architectures.
Pilot engagements for mid-market collections agencies and direct-sales companies: FDCPA/FTC compliance audits, automated QA pipeline builds, LLM evaluation harness setup.