Applied AI Engineer
Production LLM Systems · Cost Optimization · Evaluation
General Information
| Full Name | Venkata Manideep Patibandla |
| Focus | Production LLM Systems · Cost Optimization · Evaluation |
| Location | New York Metropolitan Area (open to relocation) |
| pvmanideep.analytics@gmail.com |
Summary
- Applied AI Engineer who ships LLM and ML systems to production with measured cost and quality trade-offs. Cut LLM spend 10–20× through eval-driven model routing, halved valuation error against an incumbent baseline, and deploys containerized APIs with CI-gated evals and Prometheus/Grafana observability.
Experience
- Jun 2026 - Present
AI/ML Developer
Grant Leading Technology (GLT)
- Built gIDConnect's LLM-assisted risk-assessment pipeline across four dimensions, reducing manual onboarding review time by 28% and enabling explainable admit, restrict, and block decisions.
- Evaluated model alternatives against an evaluator-curated golden dataset of 120 onboarding scenarios, selecting a configuration that reduced inference cost by 24% while maintaining 92% agreement with reviewer-approved decisions.
- Deployed scoring APIs using FastAPI and Docker; monitored production latency, error rates, cost, and sampled output quality through LangSmith traces, reducing P95 latency from 3.2 to 2.4 seconds.
- Apr 2026 - Sep 2026
AI/ML Engineer
Stealth AI Startup
- Led end-to-end development of a LightGBM automated valuation model on a 4-person team: cleaned ~88K subject-comp appraisal pairs, engineered 68+ features, and integrated RentCast API and vision-model condition scoring.
- Selected LightGBM over the incumbent comparable-property median baseline after evaluating both on a production-shaped holdout: MAPE improved from ~12% to single digits, a >50% reduction in valuation error.
- Cut error on worn-condition homes from ~24% to ~5% with vision-derived condition features, delivering the triage rule that lets the business auto-value standard stock and route only worn properties to manual appraisal.
- Apr 2025 - Jun 2026
Forward Deployed AI Engineer
SBL.so
- Shipped AI RevOps workflows across 3+ enterprise accounts (Emoha, Topmate, Clinik), automating lead research, qualification, CRM sync, and multichannel outreach; lifted outbound response rates 6–8%.
- Built lead-enrichment pipelines that fused internal and external sources into per-company intelligence, wired into client CRMs and multichannel outbound.
- Owned CRM and communication-channel integrations on the Konduit platform, resolving sync failures, lead-source gaps, and workflow errors across concurrent production rollouts.
- Oct 2022 - Dec 2023
AI Engineer & Fractional Developer
AnternData Solutions
- Delivered 5+ client-facing AI/ML applications and RAG proofs-of-concept for global technology clients including TigerData (formerly Timescale), from requirements through production deployment.
- Shipped 10+ technical demos and reference implementations across LLM APIs, retrieval pipelines, and vector-backed search to accelerate platform adoption.
Open Source Projects
- 2025 - now
CostGuard + RealDataAgentBench
- Benchmarked 12 frontier models over 1,412+ runs on 39 rubric-scored tasks across correctness, code quality, efficiency, and statistical validity, with per-run cost tracking, 95% CIs, and 196 CI tests.
- Found gpt-4.1-mini statistically tied with gpt-4.1 (0.870 vs 0.875) at 65× lower cost, then shipped CostGuard to route on it, cutting spend 10–20× vs GPT-4o defaults at ~1ms overhead, with per-provider circuit breakers, Prometheus/Grafana dashboards, and 90+ unit tests.
- 2026
MemoryOps AI
- Built and deployed a governed memory runtime (FastAPI, Postgres/pgvector, Docker/Railway) with a fail-closed production profile, tenant isolation via row-level security, and a tamper-evident audit chain.
- Gated every release on a 50-case governance benchmark at 50/50 passing with critical suites required perfect; instrumented with OpenTelemetry tracing and correlation IDs.
Technical Skills
-
LLM & Serving
- RAG, agent workflows, LangGraph, model routing and fallback, LoRA (PEFT/Unsloth), MCP
- OpenAI and Anthropic APIs, Hugging Face, Claude Code, Codex
-
ML & Evaluation
- LightGBM, scikit-learn, PyTorch, Pandas, NumPy, feature engineering
- Benchmark design, statistical validation (bootstrap/Wilson intervals)
-
Production & Observability
- Python, SQL, FastAPI, REST APIs, Docker, CI/CD, Pytest, Postgres/pgvector
- Prometheus, Grafana, OpenTelemetry, AWS (EC2/ECR/S3), Railway
Education
- Jan 2024 - May 2025
M.S. Computer Science
Sacred Heart University, Fairfield, CT
- GPA: 3.8/4.0
- Upsilon Pi Epsilon (UPE) Honor Society
- Aug 2019 - Mar 2023
B.Tech Information Technology
GMR Institute of Technology, Vizianagaram, India
- GPA: 8.30/10
Certifications
- IBM Certified in Agentic AI