Hi, I’m Livesh L, an AI Developer and final-year student at Prathyusha Engineering College. I specialize in production LLM applications, hybrid RAG with pgvector & Reciprocal Rank Fusion, in-process security firewalls, and deterministic agent reliability testing.
Automated tests & ~24.5ms RRF retrieval
In-process security & token defense
Deterministic agent tool assertions
CGPA · AI & Data Science (2027)
I am an AI developer and systems engineer focused on turning probabilistic model outputs into predictable, production-grade software. Rather than treating foundation models as black boxes or building superficial wrappers, I engineer the backend pipelines, guardrails, and evaluation harnesses that make AI dependable under real workloads.
My engineering philosophy centers on deterministic reliability over raw prompt optimism, defense-in-depth security to neutralize prompt injections and sensitive data leakage before execution, and traceable evaluation using automated regression suites and strict schema contracts.
Stateful agent loops with structured tool-calling boundaries, async streaming generators, token-budgeted context windows, and deterministic fallback paths.
Multi-tenant vector search with pgvector (HNSW) paired with sparse BM25 lexical ranking, Reciprocal Rank Fusion (RRF), and cross-encoder re-ranking.
Pre-execution prompt sanitization, heuristic injection defense, token leakage filters, and PII redaction operating locally without external latency.
Deterministic assertion suites for agent step sequences, schema contract testing with Pydantic/Zod, synthetic regression harnesses, and CI/CD quality gates.
Production AI platforms, autonomous reliability frameworks, and in-process LLM security engines backed by documented benchmarks and automated test suites.
Hybrid retrieval, context orchestration, and citation-grounded LLM synthesis for complex enterprise documents.
See hazards. Understand context. Verify evidence. Vision-based detection, deterministic safety policies, agent-assisted investigation, and interactive 3D digital-twin environments.
Autonomous incident investigation, evidence correlation, and approval-gated infrastructure remediation.
Sub-millisecond in-process firewall intercepting prompt injection, PII leaks, and rogue tool execution.
Production-grade Python libraries authored and published on the Python Package Index (PyPI). Designed for real production pipelines with strict typing, deterministic contracts, sub-millisecond execution, and automated CI/CD.
llmfirewall-core•Import:import llmfirewallLLMFirewall Core (llmfirewall-core) is published strictly as an installable in-process Python library. It is not a hosted SaaS, remote proxy, or web dashboard — all adversarial threat mitigation, entropy analysis, and Policy-as-Code checks execute natively in host process memory with sub-millisecond (+0.022ms) overhead.
A lightweight, provider-agnostic security and policy enforcement layer for LLM, RAG, and AI-agent applications. Executes in-process threat scanning to intercept adversarial prompt injections, redact sensitive credentials and PII, and enforce Policy-as-Code with sub-millisecond overhead (+0.022ms).
BLOCK > REDACT > WARN > ALLOW) with fail-closed defaults.llmfirewall CLI for CI/CD red-teaming.Zero network hop. Homoglyph decoding, invisible character normalization.
Prompt injection heuristics, credential scrubbers, PII redaction, jailbreak detectors.
Deterministic priority: BLOCK > REDACT > WARN > ALLOW.
Distribution: aireliability
An open-source reliability and regression-testing framework for LLM applications and AI agents. Replaces subjective vibe-checks with programmatic assertions across tool-invocation sequences, argument validation schemas, execution traces, and cost ceilings.
airel command-line runner for automated GitHub Actions release gates.Traces LLM outputs, function call payloads, and execution latency.
Verifies strict Pydantic argument types; rejects malformed tool calls.
Asserts valid invocation order; flags regression spikes in token consumption.
Deterministic pass. Ready for automated production release.
Non-bypassable fail-closed intercept prevents shipping broken agents.
aegis-resilience•Import:import aegisAEGIS Framework (aegis-resilience) is a lightweight, reusable open-source Python library providing foundational abstractions for incident investigation, reasoning interfaces, and tiered autonomy budgets. It is distinct from the AEGIS AI full enterprise platform.
An open-source autonomous reliability engineering framework for AI-powered incident investigation, reasoning, and remediation. Correlates multimodal observability signals across metrics, logs, and distributed traces to formulate causal failure hypotheses with non-bypassable human approval gates.
LEVEL_0 (observation) to LEVEL_3 (autonomous action).
logicfalcon•Import:import logicfalconDeveloped a modular Python code review package with 3 interfaces: CLI, SDK, and Model Context Protocol (MCP) for programmatic code analysis and developer-tool integration. Enforces safe remediation with 10 reported acceptance checks and 101/101 automated tests passed.
Automates code review with file and workspace analysis, structured findings, severity classification, diagnostic reporting, and proposed code fixes. Prepared package release artifacts for version 0.1.2 (wheel and source distribution) with strict package metadata validation and clean virtual-environment installation checks.
$ logicfalcon review ./srcfrom logicfalcon import ReviewEnginemodelcontextprotocol/v1Unlike generic wrappers that stream unverified tokens, my architectures enforce deterministic multi-stage pipelines: dense and sparse vector retrieval, in-process security firewalls, dynamic token budgeting, and open telemetry observability.
Engineered with RAG + CAG + MAG. Part of the verified multi-stage AI system architecture.
context = DynamicBudgetEngine(tier_budget=6000).allocate(rag=0.6, cag=0.3, mag=0.1)Experience production security classification and retrieval-augmented generation directly in your browser.
Every listed capability is grounded in production repositories, published packages, or verified benchmark suites.
Production retrieval systems, hybrid search, context window optimization, and grounded generation.
In-process defensive firewalls, adversarial prompt injection interception, and data loss prevention.
Deterministic testing for stochastic agents, distributed telemetry correlation, and incident recovery.
Deep learning architectures, object tracking, spatial geometry, and time-series forecasting.
High-throughput asynchronous APIs, streaming protocols, relational databases, and dynamic interfaces.
Containerization, cloud asset storage, Linux server administration, and continuous integration.
Software and machine learning systems engineered across 4 industry internships.
Engineered machine learning systems and AI evaluation pipelines for enterprise sustainability auditing and regulatory compliance.
Developed high-performance desktop clients and cloud-managed asset delivery pipelines for XR simulation modules.
Executed supervised and unsupervised machine learning workflows across raw data ingestion, feature engineering, and model validation.
Engineered business intelligence dashboards, automated data transformation models, and executive KPI reporting.
Formal engineering degree credentials and verified technical certifications.
Prathyusha Engineering College
Thiruvallur, Tamil Nadu · Aug 2023 – Expected May 2027
Core Coursework: Machine Learning, Deep Learning, Natural Language Processing, Computer Vision, Distributed Systems, Database Management, and Data Structures & Algorithms.
Grade: 100% · Certificate of Completion
Elite · 77% Consolidated Score
Verified Academic Certification
Verified Technical Certification
Professional Certificate Credential
Practical Business Intelligence
Available for full-time AI Engineering, RAG architectures, in-process security firewalls, or deterministic reliability roles starting 2026/2027.