RVReethika Valluru← Portfolio

Production AI · Systems engineering

AI Engineering
Lab

Sixteen executable systems spanning retrieval, evaluation, inference routing, caching, fine-tuning, structured generation, PromptOps, factuality, realtime voice, document AI, observability, self-hosted inference, agentic control, red-teaming, synthetic data, and AWS-native generative AI.

16production-pattern projects16interactive execution harnesses80+acceptance metrics

Retrieval-Augmented Generation

CiteRAG

Hybrid sparse–dense retrieval, reciprocal-rank fusion, cross-encoder reranking, context compression, citation alignment, and grounded answer synthesis.
Architecture
  1. Document ingestion and semantic chunking
  2. BM25 + dense ANN retrieval
  3. Reciprocal-rank fusion
  4. Cross-encoder reranking
  5. Evidence packing and citation validation
Technology stack
FastAPIQdrantOpenSearchBGE-M3CrossEncoderRAGAS

Interactive execution harness

Working demo

{
  "status": "ready",
  "action": "execute_pipeline"
}
EndpointPOST /v1/query
Example request{"query":"What is the retention policy?","top_k":5}
Response contractGrounded response with passage-level citations and retrieval scores.

Implementation guide

Build sequence

  1. 01Normalize and fingerprint source documents
  2. 02Index sparse terms and dense embeddings
  3. 03Fuse ranked candidates with RRF
  4. 04Rerank top-k passages
  5. 05Generate schema-constrained answer with citations
  6. 06Reject uncited claims

Acceptance criteria

Recall@20nDCG@10MRRFaithfulnessCitation precision

Repository

github.com/ReethikaValluru/AI-Engineering-Lab/tree/main/projects/hybrid-ragMonorepo source, contracts, tests, infrastructure, and implementation runbook.

16 Engineering invariants

Typed contracts.
Observable execution.
Measurable quality.

01

Reliability

Idempotency, bounded retries, circuit breaking, deterministic replay, durable checkpoints, and explicit failure semantics.

02

Evaluation

Golden datasets, calibrated graders, statistical thresholds, regression gates, adversarial suites, and production feedback loops.

03

Governance

Policy-as-code, provenance, audit spans, PII minimization, tenant isolation, human approval, and reversible deployment.

AI systems portfolio

From prototype
to production contract.

Every project is specified through architecture, interfaces, execution semantics, observability, evaluation, security, and deployment constraints.

Technical discussion ↗