All work
Observation log · Case ०२ · Oct 2024 → Jan 2026
Farsight
Bringing an enterprise AI ecosystem into focus.
- Client
- Omara Technologies
- My role
- Founding Full-Stack AI Engineer
- Built
- Governance portal · GT scoring · DocuNexus hub
- Stack
- Go · Python · Next.js · LangGraph · Neo4j · AWS
Scroll to look closer
I
The fog.
Farsight
Docs
Eval metrics
Permissions
Workspace A
Model scores
Labels
Gold sets
Workspace B
Runbooks
Leaderboards
Access rules
Workspace C
Docs in one place. Metrics in another.
II
The lens.
Workspace A
Docs
Models
Scores
Workspace B
Docs
Models
Scores
Workspace C
Docs
Models
Scores
Workspace D
Docs
Models
Scores
Enterprise SSO through AWS Cognito at the door.
III
Mapping the constellations.
“Which director is linked to this invoice?”
Illustrative graph
DocuNexus starts with thousands of legal and financial PDFs.
IV
Calibrating the instrument.
iAnnotateGo task engine · parallel & series
iiAgreeConsensus + expert arbitration
iiiScoreAgainst human gold standards
ivFeed backEdge cases into training
Precision
Recall
Cohen's κ
Per field, per model. Bars illustrative.
Better models came from better data, so the data got instruments.
V. What it taught me
“Don't build features for users; build intelligence engines that empower them.”
“Data is noise until it's governed by Ground Truth.”
AI wasn't a layer added at the end. Every Go service and every Neo4j schema was built to be used by agents.
Next story
AgentDiff
Agents fail quietly. A prompt tweak or model update makes the agent loop, waste tokens or cost more, and the test still passes because the final answer looks fine.
Read it