Shashank

Portfolio

A deep body of production AI work.

Eight years building software. The last several on the hardest part of AI: agent infrastructure, voice pipelines, hybrid retrieval, multi-tenant platforms, healthcare search, market intelligence and generative media each case study has the architecture written out.

All projects

The full body of work.

Nova ADK central API hub connected to Guardrails, Intent Routing, Trust Evaluation, Memory, Tool Calling and Multi-Agent modules
Agent Platform Live

Nova ADK

Full-stack platform for building and running AI agents. Intent routing, guardrails with PII detection, trust evaluation on every output, memory, tool calling and multi-agent orchestration.

Agent platform · 105+ endpoints · Public API
Case study
VectorLead Meta lead notification flows through voice waveform to qualification score 87/100 then routes to CRM
Voice AI · Sales Live

VectorLead

Inbound Meta leads called within a minute by a voice agent that qualifies in conversation, scores the outcome and routes it onward, with the calling provider swappable behind one interface.

Next.js · Supabase · LiveKit SIP · Vapi
Case study
Hybrid retrieval engine BM25 Lexical, Vector Semantic and Exact Match lanes merging through Score Fusion and Cross-Encoder Reranker into ranked results
Retrieval · RAG In production

Hybrid retrieval engine

Three retrieval channels fused in one query, cross-encoder reranking, citations verified against their sources, and an evaluation harness so quality is a number rather than an opinion.

Python · FastAPI · PostgreSQL · pgvector
Case study
Multi-tenant AI services platform 4 tenant isolation bubbles above shared AI services grid with routing, cache and token metering layers
Multi-tenant SaaS In production

Multi-tenant AI services platform

One shared substrate for routing, caching, fallback, metering and audit across ~25 AI features, each inheriting all of it rather than wiring its own infrastructure.

Multi-tenant SaaS · ~25 AI services
Case study
Enterprise platform AI CRM Live database crawled into relationship graph with Object, Field, Config nodes and schema-verified artefact output
Enterprise AI In production

Enterprise platform AI

AI over a live CRM installation. Metadata crawl into a relationship graph and configuration index, grounded answers, generated artefacts compiled and verified against the real schema before any deployment.

NestJS · pgvector · Metadata APIs
Case study
Healthcare marketplace search patient symptom query expanded through medical terminology layer into dual BM25 and dense vector search streams returning ranked specialist results
Healthcare · Search In production

Healthcare marketplace search

Patients search by symptom, not specialty. Keyword and semantic retrieval run together and are fused, with a hand-built medical terminology layer expanding the query before either sees it.

BM25 + dense · Clinical terminology
Case study
Commerce discovery and recommendations city map with scored location pins and collaborative filtering recommendations sidebar showing nearby results
Discovery · Commerce In production

Commerce discovery & recommendations

Local service discovery where geography constrains rather than competes with relevance, plus a collaborative filtering recommender and a scheduled notification engine with delivery auditing.

Semantic search · Collaborative filtering
Case study
Market intelligence engine candlestick chart with RSI and Moving Average overlays, sentiment score 0.74 Bullish, and AI analysis panel with structured output
Finance · AI Analysis In production

Market intelligence engine

Sentiment, technical indicators and strategy backtesting computed deterministically, then read by a reasoning model that writes the analysis. The model explains numbers it was given never produces one itself.

Python · FastAPI · Frontier + self-hosted models
Case study
Generative media pipeline Script generation to Speech Synthesis waveform to Video Render to Delivered, with Job State Machine, Exponential Backoff and Dead-Letter Queue below
Generative Media In production

Generative media pipeline

Script to synthesised speech to rendered video, built around job state, exponential backoff, duplicate prevention and a dead-letter path because generative media APIs fail more than their docs admit.

Script · speech · video · provider adapter
Case study

Also built

More from the archive.

Search & recommendations

  • Semantic search with geographic ranking
  • Collaborative filtering recommender
  • Demand tracking
  • Scheduled notification engine with delivery audit

Document intelligence

  • Extraction from PDF and Office formats
  • Schema-constrained field capture
  • Validation and confidence routing
  • Form completion and PDF filling

Generative media

  • Script to synthesised speech to video
  • Provider adapter for voice engines
  • Retry with exponential backoff
  • Duplicate prevention and dead-letter handling

Enterprise platform AI

  • CRM metadata analysis at scale
  • Requirement document generation
  • Configuration and code generation
  • Deployment with preview and rollback