17 projects across agentic AI, RAG/LLM, classic ML, and the Go/Rust/TypeScript backends & DevOps that ship them — all live, not just prototyped.
flagship.01
TREEHASH
LIVE
treehash — Vectorless RAG Engine
Python · SQLite (FTS5) · PostgreSQL · BM25 · FastAPI · MCP · Docker
- Parses documents into a hashed address tree and resolves most queries by direct lookup — no embeddings, no vector index, O(1) average latency.
- Falls back to BM25 full text search only when structural resolution misses, with an optional semantic tier gated behind both failing first.
- Cut parsing/indexing stage timeouts by 95% by tracing and fixing a page cache memory leak in the PDF parser, holding ingestion stable at 100MB+/multi document scale.
- Hash chained audit trail and real provider billed token accounting per query; published to PyPI as
treehash-rag.
Zero Vector DB92% Accuracy95% Fewer TimeoutsPyPI Published
flagship.02
AP AUTOPILOT
LIVE
AP Autopilot — Accounts Payable Automation
Temporal · Go · PostgreSQL · Next.js · MCP · QuickBooks/Xero/NetSuite/Stripe
- Separates reasoning from authority — LLMs extract invoice data only; a deterministic Go policy gate makes every payment decision.
- 94.2% field extraction accuracy with a 0% false approval rate; auto approves low risk invoices in ~0.5s, escalates the rest to named approvers.
- Hash chained, replayable audit ledger on durable Temporal workflows that survive restarts without losing state.
0% False Approval94.2% ExtractionBounded Autonomy
flagship.03
VERIFACT AI
LIVE
VeriFact AI — Zero Hallucination RAG
Hybrid Retrieval · Multi Tenant Isolation · Encryption at Rest · MIT Core
- Answers strictly from the user's own documents — a verbatim quote or an honest "I don't know," never an invented answer.
- Confidence scored hybrid retrieval benchmarked on 94 real questions, with per user isolation and encryption at rest for multi tenant use.
- Modular, MIT licensed retrieval core kept cleanly separable from the product shell.
Zero Hallucination94 Q BenchmarkMulti Tenant
flagship.04
SENTRYOPS
LIVE
Sentryops — Autonomous DevOps Agent
Rust · Go SDK · Kubernetes · OPA style Policy · Groq · MCP
- Discovers Kubernetes pods, runs multi endpoint health checks, and auto files tickets the moment a service degrades.
- Clusters raw logs into causal dependency graphs, then an LLM diagnoses root cause and proposes remediation — ranked by actual cause, not symptom.
- Auto resolves known safe errors behind an OPA style default deny policy gate, with correlation ID audit trails on every decision.
Rust CoreCausal RCADefault Deny Gate
flagship.05
LEGAL CONTRACT PIPELINE
LIVE
Legal Contract Pipeline
LangGraph · FastAPI · Pinecone · PostgreSQL · Celery/Redis · React · Jenkins · MLflow · Kubernetes
- Automated first pass legal review via a 5 agent workflow (Extractor, Scorer, Compliance Checker, Redliner, Writer).
- Grounded LLM risk assessment in precedent — RAG with Pinecone retrieves top 3 similar historical clauses.
- Gated deployments on a 60%+ calibration accuracy threshold via a Jenkins + MLflow eval harness.
5 Agent Workflow60%+ Eval GateRAG Top 3
flagship.06
DEVOPS HELPDESK AGENT
LIVE
Autonomous DevOps Helpdesk Agent
Python · Agentic Workflows · FastAPI · SQLite · Vector Store · Kubernetes · Jenkins
- Cut incident response time with an autonomous agent that plans, calls tools, and reasons over runbooks and live alerts.
- Tiered safety guard system — auto approves read only calls, routes state mutating actions to manual sign off.
- Full audit traceability via an immutable, append only Postgres audit log.
Human in the LoopParallel ExecImmutable Audit Log
flagship.07
DEVINTEL
LIVE
DevIntel — Competitive Intelligence Pipeline
FastAPI · Next.js · PostgreSQL · Redis · Playwright · Claude/GPT 4o/Groq · Docker
- Full stack SaaS that monitors competitor URLs on a schedule, detects changes via SHA 256 hashing, and generates LLM intelligence reports.
- Multi channel report delivery (Slack, Email, Webhook) via Redis/RQ async jobs and APScheduler cron.
- Estimated 3–5 hours/week saved per engineering team through end to end automation.
SHA 256 Diffing3–5 hrs/wk savedMulti channel
flagship.08
INTELLIASSIST
RESEARCH
IntelliAssist — RAG Enhanced Virtual Assistant
PyTorch · Hugging Face · FAISS · LoRA · Whisper · Gradio · Docker
- 3.4× context relevance over a vanilla RAG baseline via hybrid FAISS + BM25 search, MMR ranking, and cross encoder reranking.
- LoRA adapters on GPT 2 cut fine tuning params by 95%, hitting a 96% pass rate across 25 eval cases.
- Sub 200ms average response latency with a multimodal interface and hallucination detection fallback.
3.4× Relevance96% Pass Rate95% Param Cut
flagship.09
MODELFORGE
LIVE
ModelForge — Agentic Model Remediation
Express/TS · Go · Rust · MongoDB/PostgreSQL/S3 · OpenAI/Claude/Gemini (BYOK)
- Multi agent pipeline (QA → Diagnose → Research → Strategize) reasons over real computed metrics — every agent response is schema validated before Go dispatches it; no agent ever emits code that runs.
- Trains real remediation candidates in Rust (reweighing, resampling, hyperparameter search, neural net, RAG grounding) and picks a winner via Pareto frontier selection on identical held out rows.
- Verified run: 44.4% → 94.4% accuracy (+50pp) on an 18 row held out split — same rows scored before and after, no cherry picking, no mocked numbers.
Real Rust Retrains+50pp VerifiedSchema Guardrailed Agents
IntelliAssist — RAG AI Assistant
Production RAG system on fine tuned GPT 2 + LoRA. 96% pass rate across 25 test cases, 0.15s avg response, hybrid FAISS+BM25 search, multimodal I/O.
PythonLoRAFAISSWhisperDocker
PaperIntel AI — Research Paper Analyzer
Multi agent LangGraph platform that summarizes papers, flags methodology issues, and produces PPTX presentations. FastAPI + PostgreSQL + Redis.
LangGraphFastAPIGroqK8s
Indic NLP Microservice
Multilingual sentiment (5 star BERT) + zero shot topic classification (mDeBERTa) for English, Hindi, Hinglish. Async batch processing, K8s + HPA.
BERTmDeBERTaK8sAWS
HireSparkAI — Talent Intelligence
Dual RAG pipelines for freshers vs experienced hires. Explainable matching, interview question generation, resume optimization. ArgoCD + Jenkins CI/CD.
FAISSLangChainArgoCD
Secure MLOps Recommendation API
Neural Collaborative Filtering engine (MovieLens 100k) with DevSecOps hardening — Bandit, Trivy, Dockle, Cosign signing, MLflow tracking.
PyTorchMLflowTrivyK8s
LangChain Chain Patterns
10 working LangChain chain pattern implementations — sequential, router, memory, tool use — each with docs and graph visualizations.
LangChainGroqPython
DocuScribe — AI Document Processor
Extracts structured insights from PDFs/images via Claude vision API, generates SRT subtitles, auto tags docs. Handles 100+ docs/day.
Claude VisionCeleryRedis
TurboLearn — Adaptive Learning
AI personalized learning paths with SM 2 spaced repetition, AI tutoring chat, and 2,000+ concurrent users at sub 200ms latency.
Next.jsFastAPIPostgreSQL
DevIntel — Competitive Intelligence
Schedules competitor URL monitoring, SHA 256 diffing, LLM report generation, and multi channel delivery via Slack/email/webhook.
Next.jsPlaywrightRedis
DevOps Helpdesk Autonomous Agent
Parses infra alerts, reasons over conflicts, executes safe actions behind a human approval gate, with full ID correlated audit logs.
FastAPIVector StoreGroq
Legal Contract Pipeline
Multi agent legal compliance pipeline via LangGraph, Pinecone risk calibration, and a 60% accuracy eval gate blocking bad merges.
LangGraphPineconeMLflow
Python Q&A Assistant
RAG system grounded in 50k Stack Overflow pairs — MiniLM embeddings, ChromaDB retrieval, Groq generation, ~700ms avg latency.
ChromaDBSentence TransformersGroq
treehash — Vectorless RAG Engine
Parses documents into a hashed address tree, resolving most queries by O(1) average lookup instead of vector search. Cut parsing/indexing stage timeouts by 95% via a page cache memory leak fix; BM25 fallback, hash chained audit trail, published on PyPI.
PythonSQLite FTS5PostgreSQLMCP
AP Autopilot — AP Automation
Invoice extraction, matching, and risk scoring with a deterministic Go policy gate that alone can approve payment. 94.2% extraction accuracy, 0% false approvals.
TemporalGoPostgreSQLNext.js
VeriFact AI — Zero Hallucination RAG
Answers strictly from a user's own documents with confidence scored hybrid retrieval — a verbatim quote or an honest "I don't know," benchmarked on 94 real questions.
Hybrid RetrievalMulti TenantMIT Core
Sentryops — Autonomous DevOps Agent
Discovers K8s pods, clusters logs into causal dependency graphs, and diagnoses root cause via LLM, auto resolving known safe errors behind a default deny policy gate.
RustKubernetesOPA PolicyGroq
ModelForge — Agentic Model Remediation
Multi agent pipeline diagnoses an underperforming model, trains real Rust remediation candidates, and picks a winner by Pareto frontier selection — verified 44.4% → 94.4% on held out data.
GoRustExpress/TSMulti Agent